What is the Kelly criterion?
A bankroll-growth formula that ties bet size to your actual edge โ powerful in theory, dangerous at full strength, and only as good as your probability estimate.
The Kelly criterion is a formula that tells you what fraction of your bankroll to bet to maximize long-run growth, given your edge and the odds. It stakes more when your advantage is larger and less when it is thin, and mathematically it grows a bankroll faster than any other strategy โ but only if your win-probability estimate is accurate. Because real estimates are noisy, most practitioners bet a fraction of full Kelly to reduce swings.
For even-money-style bets, Kelly fraction = edge รท odds, or more generally f = (bp โ q) รท b, where b is the decimal odds minus one, p your win probability and q = 1โp. If a โ110 bet (b โ 0.91) has a true 55% chance, f โ (0.91ร0.55 โ 0.45) รท 0.91 โ 5.5% of bankroll. That is a large bet, which is exactly why full Kelly is volatile.
Full Kelly assumes you know your edge exactly. You don't. Overestimate your probability โ easy to do โ and Kelly tells you to bet far too much, risking ruin. Half- or quarter-Kelly keeps most of the growth with a fraction of the swings. It also connects directly to CLV: if you can't show you beat the closing line, your edge estimate is suspect, and Kelly-sizing on a phantom edge is how bankrolls die.
What is the Kelly criterion?It is a formula for sizing bets as a fraction of bankroll to maximize long-run growth, betting more when your edge is larger. It grows money fastest in theory but only if your win-probability estimate is correct.
Why do people use fractional Kelly?Full Kelly is extremely volatile and punishing when you overestimate your edge. Betting a half or quarter of the Kelly amount keeps most of the growth while sharply reducing the risk of large drawdowns.
Is Kelly safe for beginners?Not at full strength. Because it depends entirely on an accurate edge estimate โ which beginners rarely have โ flat unit staking or a small Kelly fraction is far safer until you can prove your edge, ideally through closing line value.