Kelly Criterion Calculator for Sports Betting

Kelly turns two numbers into a stake: the price the bookmaker offers and the probability you believe the bet has. The calculator above starts on decimal odds of 2.10 with an estimate of 55 percent, and full Kelly comes out at 14.09 percent of the bankroll, just under a seventh of it on one bet.

Treat that as an upper limit. How close to it you go depends on how far you trust that 55 percent.

How much of the bankroll to stake

The norm is to bet a fraction of full Kelly, because the full stake rests entirely on an estimate you made yourself. Get the estimate exactly right and full Kelly grows the bankroll faster than any other fixed rule. Get it slightly wrong and it grows slower, with swings that most people cannot sit through.

• Full Kelly, 14.09 percent here: the fastest growth if your estimate is exactly right, and the deepest losing runs.
• Half Kelly, 7.05 percent: gives up a little growth on paper for much smaller swings.
• Quarter Kelly, 3.52 percent: the usual default for anyone whose probabilities do not come from a tested model.

The trade-off is worth it because the estimate is always off by something, and a fraction of a stake that was too large is closer to right than the whole of it. The calculator also asks for a risk ceiling, a share of the bankroll you will not go past whatever the formula says. It starts at five percent, and the final stake is whichever is lower: the fraction you chose or the ceiling.

What the calculator needs from you

The decimal odds come from the bookmaker and are a fact. The win probability comes from you, and it is the fragile half: nobody hands it to you, and nothing on the betslip tells you whether it is right.

From those two the calculator reads off two more numbers before it sizes anything. The break-even probability is the chance at which the price is fair, here 47.62 percent. The edge is what you expect to make per unit staked if your estimate is right, here 15.5 percent. The bigger the edge, the bigger the stake, and the same edge at a shorter price asks for a bigger stake than at a longer one.

The calculator takes decimal, American and fractional odds directly, and it will accept your estimate as a percentage or as the fair odds you think the bet deserves. Use the odds converter if you want to see a price in every format at once, and read the guide to positive expected value betting first if you are not yet sure an edge exists at all. Kelly sizes an edge; it does not find one.

When the answer is zero

Zero is a real answer, and the calculator returns it whenever the estimate sits at or below break-even. A negative result means no bet, not a bet on the other side. There are also times to take zero even when the formula says otherwise:

• the estimate has no track record behind it
• the settlement rules of the market are unclear
• the price has moved since you worked out the edge
• the bankroll is money you cannot afford to lose
• the bet overlaps with others you already hold and you cannot say by how much

How wrong your probability can be

The formula never changes; the number you feed it does. At a fixed price the stake rises in a straight line with the probability you type in, and the line is steep. Knock the estimate down from 55 percent to 47, just under the break-even of 47.62, and the answer is no bet at all. Eight points of a number you guessed separate no bet from about a seventh of the bankroll.

No one can show that a single-match estimate is right to within eight points. That is why the fraction matters more than the formula: a quarter of a stake that was twice too large is still a stake you can afford. It is also why a range is more honest than a single number, and why the stake should come from the cautious end of it.

Before you trust your own number

Kelly's most important input is the one nobody checks. Before you size anything with it, ask whether your probabilities have earned the right to be used:

• Have you written down estimates before the result on a meaningful number of bets, and compared them with what happened?
• When you called a group of bets likely at a given rate, did they win at about that rate?
• Does the record hold up market by market, or does one sport carry the rest?
• Is the estimate the one you would have made before seeing the price, or did the price nudge it?

No record means the formula's key input is unsupported, and the honest answer is zero or a small fixed research stake while you build one. If your central estimate clears break-even but the cautious end of your range does not, the answer is also zero. Write down why, so the optimistic number does not quietly replace the rule you set.

Correlated bets and moving odds

Kelly sizes one bet at a time and assumes the others are unrelated. Two bets that each take a quarter-Kelly stake look like two small positions, but if both depend on the same team, the same injury or the same weather, a bad night hits both at once and you are holding one large position.

• Group open bets by what they share: a team, a player's fitness, the weather, the referee.
• Cap the total stake on each group, then size each bet inside the cap.
• Treat a parlay and its own legs as one bet counted twice, not as separate positions.

Moving odds need the same care. A Kelly stake belongs to one probability and one price: if the price has shortened by the time you reach the betslip, the edge has shrunk with it, so work the stake out again. There are two smaller adjustments. A market that can push has three outcomes rather than two, so the two-outcome formula does not fit it exactly. On an exchange, take the commission off the winning payout before you enter the odds.

Update the bankroll after every settlement

Kelly is a fraction of the bankroll you have now, not the one you started with. After a loss the next stake is worked out from the smaller figure; after a win, from the larger one. That is what softens losing runs: the stake shrinks as the bankroll does. It does not remove them. Five losses in a row at full Kelly on this page's example would leave less than half of the bankroll, and that is an ordinary bad run, not a freak one.

• Work the next stake out from the settled bankroll, after every result.
• Do not count unsettled bets as money you have.
• Keep cash for open liabilities and ordinary life outside the bankroll entirely.
• Do not top the bankroll back up in the middle of a losing run; a deposit hides what the strategy did and spoils any comparison you make later.

The formula, if you want it

The working below uses the net odds: what a winning unit stake pays on top of itself.

Kelly picks the fraction that makes the bankroll grow fastest over a long run of bets with known probabilities. It does not maximize the chance of winning the next bet, it does not limit losing runs, and it says nothing about a single season. Reading the output as a prediction is the most common way to misuse it.

Frequently asked questions

Sources checked

• J. L. Kelly Jr., "A New Interpretation of Information Rate", Bell System Technical Journal (1956) (the original formula; link checked 19 September 2026)
• The calculator's Kelly formula, checked against reference values by automated tests (run 19 September 2026)