NBA
MLBKelly Staking
The math says bet 5% of your bankroll. The sportsbook says your account is suspended. Kelly staking is the rule for turning an edge into a bet size — and knowing it cold is the difference between a roll that compounds and one that taps out on a bad week.
What Kelly staking is
Kelly staking is bankroll-percentage sizing scaled to your edge and your odds.
Bet too little and you leave growth on the table; bet too much and a normal cold streak ruins you before the edge pays off. The Kelly criterion solves for the single bet fraction that maximizes long-run growth of the bankroll — the mathematically optimal middle. It does not care how confident you feel. It cares about exactly two things: how likely you are to win, and how much you collect when you do.
Kelly is not a betting system that finds winners — it sizes bets you have already established as +EV. No edge, no stake. Find the edge first with expected value and de-vigging; then Kelly tells you how big.
The formula
f* = (b × p − q) ÷ bThree inputs, each plain-English:
- pyour true win probability — your model's honest estimate, not the book's implied number.
- qthe probability you lose, which is simply 1 − p.
- bthe net decimal payout — what you win per 1 staked. A +120 underdog pays b = 1.2; a −110 favorite pays b ≈ 0.909.
The numerator b × p − q is your edge expressed in payout terms. When it is positive, you have a bet worth making; when it is zero or negative, Kelly stakes nothing — the formula passes for you.
Put a number on it
Drag your win probability, pick the offered price, and watch the formula resolve. Notice how a thin edge produces a small full-Kelly stake — and how, the moment your probability drops to the book's implied number, Kelly drops to zero.
Quarter Kelly is the engine's default stake guidance — most of the long-run growth, a fraction of the swings.
Why sharps bet a fraction of Kelly
Full Kelly is optimal only if your edge estimate is perfect — and it never is.
The formula assumes you know your true win probability exactly. In reality that number is a forecast with error bars, and full Kelly punishes overconfidence brutally: bet full Kelly on an edge you have overestimated by a few points and you are wildly overstaked, riding a curve that swings from euphoric to ruinous on variance alone. Halving the stake roughly quarters the volatility while keeping about three-quarters of the long-run growth. That trade is why serious bettors almost never bet more than half Kelly, and many sit at a quarter.
Dynatyze's +EV engine publishes a quarter-Kelly stake on every surfaced pick by default — built for the reality that your edge estimate is uncertain and your account is mortal, not for a textbook where neither is true.
Bankroll stages
The right fraction is not just about variance tolerance — it is about what your bankroll and your books will actually let you do. As the roll grows, the binding constraint shifts from survival to book limits and account longevity.
Small bankrolls feel every swing. Quarter Kelly keeps a cold streak from gutting the roll before the edge has a chance to show, and it keeps single-bet limits at recreational books from forcing you off-size.
With more cushion you can lean toward half Kelly on your highest-confidence edges while holding quarter on the marginal ones. Spread across more books and the per-bet limit stops dictating your stake.
At scale the binding constraint is rarely the math — it is book limits and account longevity. Half Kelly across many books captures growth while keeping any one account from drawing a limit before you have placed the bet.
What 300 picks looks like
An edge is not a payday; it is a slope. Below is the textbook expected-growth curve for a modest, realistic edge — a 55% win probability on a −110 price — staked at full, half, and quarter Kelly across 300 settled bets. These are expected-value curves from the growth formula, not a simulation and not anyone's real results.
Log scale. Textbook expected-growth curves, not a simulation or real results.
Full Kelly grows fastest on paper — and swings hardest in practice, because a single overestimate of your edge turns the same curve into a drawdown. Fractional Kelly trades a sliver of expected growth for a far smoother ride.
Two lessons. First, even a small edge compounds meaningfully over a few hundred plays — which is why 300 settled bets is the rough horizon where a real edge separates from noise. Second, the steepest paper curve is the most dangerous one to ride, because the model assumes your edge is exactly what you think it is. Smooth and certain beats steep and fragile.
Kelly, answered
Yes — Kelly staking is a bankroll-sizing rule, not a way to beat a sportsbook, so it is entirely legal to use. The practical catch is that books may limit or close accounts that consistently bet winning numbers, regardless of how you size them, because winning is what triggers a review, not the staking math.