What Kelly actually solves
Bet too little and your edge barely compounds. Bet too much and a normal losing streak wipes you out before the edge can pay off. Somewhere between those two extremes is a single fraction that grows capital faster than any other over a long series of bets. That fraction is what the Kelly Criterion computes.
It was published in 1956 by John Kelly, a researcher at Bell Labs, and later adopted by professional gamblers and investors. The idea is not to maximize the return of any one bet, it is to maximize the long-run growth rate of the whole account. Those are different goals, and the difference is exactly why Kelly refuses to let you bet the farm even on a strong hand.
The formula
- f* = the fraction of capital to risk on the bet.
- W = your probability of winning, as a decimal. A 55% win rate is 0.55.
- R = the payoff ratio, your average win divided by your average loss. Wins twice the size of losses is 2.0.
Read it in plain language and it is intuitive. The first term rewards you for winning often. The second term penalizes you for losing, scaled down by how much bigger your wins are than your losses. A high win rate paired with fat wins pushes the fraction up. A thin edge pushes it toward zero, and a losing strategy drives it negative, which is the formula telling you to stand aside.
A worked example (and why it is scary)
Suppose you win 55% of your trades, so W = 0.55, and your winners average twice your losers, so R = 2.0. Drop those into the formula:
- Losing term = (1 − 0.55) ÷ 2.0 = 0.45 ÷ 2.0 = 0.225.
- Kelly fraction = 0.55 − 0.225 = 0.325.
- Full Kelly says risk 32.5% of your capital on this one trade.
Read that number again. Full Kelly, from a perfectly reasonable-looking edge, tells you to put nearly a third of your account at risk on a single position. No disciplined trader does that. A short run of losses at 32.5% per trade carves the account to pieces, and it only takes a few in a row to dig a hole that math alone cannot climb out of. The formula is not wrong. It is answering a question about infinite, perfectly known bets, and you do not live in that world.
Why half-Kelly, or even quarter-Kelly
The fix that professionals reach for is fractional Kelly: take the number the formula gives you and bet a fraction of it. Half-Kelly is the common choice, and the tradeoff behind it is remarkable. Cutting the bet to half Kelly keeps roughly three-quarters of the growth rate while cutting the volatility in half and making the drawdowns dramatically shallower. You give up a little compounding and you buy a much smoother ride, one you can actually stay on.
In our example, half-Kelly turns 32.5% into 16.25%, and quarter-Kelly into about 8%. Both are still aggressive by most standards, which tells you how hot the raw number really was. The deeper reason to cut it is that your W and your R are not facts, they are estimates pulled from a limited trade history. Bet full Kelly on an edge you have overestimated and you are not sizing to your real advantage, you are sizing past it, straight into the ruinous drawdowns Kelly was supposed to help you avoid. Betting a fraction builds in a margin for being wrong about yourself.
The assumptions Kelly makes, and how real trading breaks them
Kelly leans on three assumptions, and live markets bend all three. It assumes you know your edge precisely, when in truth every win rate and payoff ratio you have is a noisy sample that shifts as conditions change. It assumes each outcome is independent, when correlated positions and changing regimes tie your trades together. And it assumes the bet is repeatable on identical terms, when no two setups are ever exactly the same.
None of this makes Kelly useless. It makes Kelly a ceiling rather than a target. The number it produces is the most you could ever justify betting if the world were perfectly knowable, and since it is not, you sit well below that ceiling on purpose. Kelly tells you which direction is too much; fractional Kelly keeps you a safe distance from it.
One more connection worth drawing: Kelly answers “what fraction,” not “how many shares.” Once you have your fraction, a position-size calculator turns it into an actual order by combining it with your entry and your stop. Kelly sets the risk budget, and the sizing math spends it.