Kelly Criterion for Sports Betting in 2026: The Formula, Fractional Kelly and Estimate Error

The quick answer

• Formula: f* = (b x p - (1 - p)) / b, where b is the net decimal odds (decimal price minus 1) and p is your win probability.

• Worked example: at decimal 2.10 with p = 50%, f* = 4.55% of bankroll.

• Sensitivity: at 2.10, every point of p moves the stake by 1.91 points of bankroll. At 52% the formula says 8.36%; at 48% it says 0.73%.

• Overbetting: at twice the Kelly fraction, long-run growth falls to zero.

• Fractional Kelly: at 2.10 and a true 50%, half Kelly keeps 75% of the growth rate at half the stake.

• Fees: the vig or fee lowers b, and a lower b cuts the stake. At a 53% estimate, Kelly says 1.30% at -110 and 6.00% at a 50% price with no fee, as on a Novig pre-game straight.

The Kelly criterion gives the share of your bankroll that maximizes long-run growth, provided you know the true probability. You never do in sports betting, and the formula reacts so strongly to small errors in p that full Kelly on an optimistic estimate overbets.

What the Kelly criterion is

John Kelly published it in the Bell System Technical Journal in July 1956, in a paper on information theory titled "A New Interpretation of Information Rate." His setup is a gambler with a private wire who learns results, imperfectly, before the odds change.

Kelly showed why betting everything on each favorable bet doesn't work. It maximizes expected capital, but the gambler "would be broke with probability one if he continued indefinitely." Betting a fixed fraction instead, he found the fraction that maximizes the growth rate of capital, and showed that a gambler using it "will, with probability one, eventually get ahead and stay ahead of one using any other" fixed fraction.

In his even-money case that fraction is the win probability minus the loss probability. The general version for any price is the formula above.

How we checked

Every figure below is computed from the formula and from the growth rate G = p x ln(1 + b x f) + (1 - p) x ln(1 - f), with assumptions stated. Kelly's quotes come from the 1956 paper, which we read on 7 October 2026. Novig's fees come from its fee page, checked the same day. Prices in the examples are illustrative.

How to calculate Kelly: worked examples

Step 1, convert to net odds. b = decimal price minus 1. 2.10 gives b = 1.10. -110 is 1.9091, so b = 0.9091.

Step 2, plug in your probability. At 2.10 and p = 50%: f* = (1.10 x 0.50 - 0.50) / 1.10 = 0.05 / 1.10 = 4.55%.

Step 3, stake that share of your current bankroll, and recalculate as the bankroll changes.

The first two rows carry the same +5% edge and get different stakes. Kelly works out to EV divided by b, so the same edge on a short-priced favorite earns a bigger fraction than on an underdog, where the variance is higher. A negative result means you don't bet.

How sensitive is Kelly to a probability error?

The stake moves by (b + 1) / b points of bankroll for every point of p. At 2.10 that's 1.91 points; at -110 it's 2.1.

A two-point swing in your estimate changes the stake from 0.73% to 8.36%, more than 11 times as much.

Say you believe 52% at 2.10 and the truth is 50%. Full Kelly on your belief stakes 8.36%, which is 1.84 times the correct 4.55%, and your long-run growth rate falls to 29% of what correct staking would earn. Half of your believed Kelly, 4.18%, earns 99.4% of it.

If the truth is 48% and you think 50%, the bet has negative EV. Full Kelly on your estimate shrinks your bankroll at about 0.077% per bet in log terms, and half Kelly at about 0.010%. Both lose money; half Kelly loses far less.

What fractional Kelly costs and saves

Growth at different multiples of the correct Kelly fraction, at 2.10 with a true p of 50%:

In this example growth is symmetric around full Kelly: 0.5x and 1.5x earn the same growth, and the 1.5x bettor carries three times the stake to get it. Past 2x, a bettor with a true edge loses money over time.

Half Kelly gives up a quarter of the growth rate for half the stake per bet. When p is an estimate the true Kelly fraction isn't known, and a bettor at full Kelly on an optimistic estimate is already somewhere past 1x, which is the case the error example above prices out.

Why the Kelly fraction is a ceiling for one bet

Kelly's paper names "the possibility of reinvestment of profits and the ability to control or vary the amount of money invested" as "essential requirements for the validity of the theory." The math also treats p as known and each bet as independent.

Two of those assumptions fail in sports betting:

• Unknown p: your probability is an estimate, and the stake inherits its error at about two points per point of p.

• Correlated bets: a side and a total on the same game, or two props on one player, depend on the same outcome. Sizing each at full Kelly as if they didn't depend on each other stacks the exposure.

So treat f* as an upper bound for one bet, scale it down for estimate error, and cap total exposure across open positions on the same game.

How fees change b

Kelly's b is the net odds you get after costs, and vig and fees both shrink it.

Say your estimate on an NFL side is 53%:

The same read sizes at 1.30% of bankroll against -110 and 6.00% at an even price with nothing added. The 50% price is an assumption for the example and doesn't come from any venue's board.

On Novig, pre-game straight trades carry no fee for maker or taker, so b is the price itself. Novig's fee page says "There's no vig or juice built into our prices." Live straights carry a taker fee of 0.03 x price x (1 - price), which is the third row. Futures carry 0.06 (golf and tennis futures carry none as of 10 September 2026), parlays carry 0.10 built into the price, and makers pay nothing on any trade type.

What to look for

• Net odds: compute b from the price after the vig or fee on the exact bet type. Where a pre-game price has no vig or fee added, b is the price itself.

• Estimate error: decide how many points your p could be off and size for the low end.

• Fraction: start at a quarter to half Kelly when p comes from an estimate.

• Exposure: add up every open position that depends on the same game before sizing the next one.

• Bankroll: recalculate each stake from your current bankroll.

Kelly criterion FAQ

What is the Kelly criterion formula for sports betting?

f* = (b x p - (1 - p)) / b, where b is the decimal price minus 1 and p is your win probability. At 2.10 and 50%, f* = 4.55% of bankroll.

Is full Kelly too aggressive for sports betting?

Usually, because p is an estimate. At 2.10, overestimating p by two points makes full Kelly stake 1.84 times the correct amount and cuts long-run growth to 29% of the maximum.

What is half Kelly?

Staking half the Kelly fraction. At 2.10 with a known 50% chance, it keeps 75% of the long-run growth rate at half the stake.

What happens if you bet more than Kelly?

Growth falls, and at twice the Kelly fraction it's zero. Beyond that, a bettor with a true edge still loses money over time.

Do betting fees change the Kelly stake?

Yes. Fees lower the net odds b, which lowers the stake. A 53% estimate sizes at 1.30% of bankroll at -110 and 6.00% at a 50% price with no fee.

Sources

All checked 7 October 2026.

Kelly, J. L., Jr., "A New Interpretation of Information Rate," Bell System Technical Journal 35(4), July 1956, 917-926: archive.org/details/bstj35-4-917

Novig fee schedule: support.novig.com/en/articles/16195057-fees-on-novig