The Kelly Criterion tells you what share of your bankroll to stake on a bet, based on your edge and the odds. The formula is: stake = (your probability x decimal odds - 1) / (decimal odds - 1). If you think a team wins 55% of the time and the price is -110 (1.91 in decimal odds), Kelly says to stake 5.5% of your bankroll.
That is the math. The practical answer is smaller. Most bettors should stake a quarter of the Kelly number, because Kelly assumes your probability is exactly right, and it never is.
Below is how to calculate it, why full Kelly is too aggressive, and what fraction to use instead.
What is the Kelly Criterion in sports betting?
It is a staking formula. John Kelly, a scientist at Bell Labs, published it in 1956, and it gives the stake size that grows a bankroll fastest over a long run of bets.
It does two useful things for a bettor. It stakes more when your edge is bigger, and it stakes nothing when you have no edge at all. It also keeps every stake as a share of your current bankroll, so stakes grow when you win and shrink when you lose.
What it does not do is find the edge for you. Kelly needs one input that only you can supply: your own probability for the bet.
How do you calculate a Kelly stake?
You need two numbers: your probability that the bet wins, and the decimal odds.
Kelly stake = (probability x decimal odds - 1) / (decimal odds - 1)
The top part of the formula is your edge, the expected profit per dollar staked. The bottom part is the net odds, what you win per dollar if the bet wins.
Here is the favorite example from the top:
- Your probability: 55%
- Price: -110, which is 1.91 in decimal odds
- Edge: 0.55 x 1.91 - 1 = 0.05, so +5% per dollar staked
- Kelly stake: 0.05 / 0.91 = 5.5% of your bankroll
With a $1,000 bankroll, that is a $55 stake.
In Google Sheets, put your probability in B2 and the decimal odds in C2, then use =MAX(0,(B2*C2-1)/(C2-1)). The MAX part returns zero when there is no edge, so the sheet never tells you to stake on a bad price.
If you only have American odds, convert them first. A negative price like -110 becomes 1 + 100/110 = 1.91. A positive price like +250 becomes 1 + 250/100 = 3.50.
Why does Kelly stake less on underdogs?
Because the same edge comes with much more variance on a longshot.
Take an underdog with the same 5% edge:
- Your probability: 30%
- Price: +250, which is 3.50 in decimal odds
- Edge: 0.30 x 3.50 - 1 = 0.05, so +5% again
- Kelly stake: 0.05 / 2.50 = 2.0% of your bankroll
| Bet | Your probability | Price | Edge | Kelly stake |
|---|---|---|---|---|
| Favorite | 55% | -110 (1.91) | +5% | 5.5% |
| Underdog | 30% | +250 (3.50) | +5% | 2.0% |
Same edge, less than half the stake. The underdog loses 7 times out of 10, so long losing runs are normal, and Kelly sizes the bet so those runs do not sink you. The longshot bets lesson shows how fast a bankroll can drain when longshots are staked like favorites.
Why is full Kelly too aggressive for most bettors?
Because the swings are bigger than almost anyone can live with, even when your edge is real.
Here is a simulation: 20,000 runs of 1,000 bets each at -110 (1.91), with a true 55% win rate, starting from the same bankroll, at three stake sizes:
| Stake size | Stake per bet | Typical bankroll after 1,000 bets | Chance the bankroll halves at some point |
|---|---|---|---|
| Full Kelly | 5.5% | 4.0x | 43% |
| Half Kelly | 2.75% | 2.8x | 10% |
| Quarter Kelly | 1.4% | 1.8x | 0.4% |
Full Kelly grows the bankroll fastest. It also cuts the bankroll in half at some point in more than 4 out of 10 runs, with a real 5% edge on every bet. Most people stop following a system long before that.
Quarter Kelly still grows the bankroll by about 80% over those 1,000 bets, and the chance of ever losing half of it is close to zero. You give up some growth and you get a plan you can actually stick to.
There is also a hard ceiling. At twice the Kelly stake, expected growth falls to about zero, even with the edge. Above that, you lose money in the long run while winning more than half your bets. The bankroll growth lesson lets you drag the stake size and watch this happen.
What happens if your probability is wrong?
Full Kelly turns a small mistake into a losing bankroll.
Say your model says 55%, but the true probability is 53%. At -110 you still have an edge, about +1.2% per dollar. Here is what the same three stake sizes do over 1,000 bets when you size them for 55% but the truth is 53%:
| Stake size | Typical bankroll if the truth is 55% | Typical bankroll if the truth is 53% |
|---|---|---|
| Full Kelly | 4.0x | 0.48x |
| Half Kelly | 2.8x | 0.98x |
| Quarter Kelly | 1.8x | 1.08x |
With full Kelly you had an edge on every bet and still lost half your money. Quarter Kelly still made a profit with the same wrong number. That is the main reason to use a fraction: it protects you from your own probability.
And your probability is always a little wrong. My NFL model went 146-112 (56.59%) over the last five seasons, with positive closing line value against Pinnacle's closing price. If you put 56.59% into Kelly at -110, it says to stake 8.8% on every bet. But 258 bets cannot pin down a true win rate that precisely. The range that fits that record at 95% confidence runs from about 50.5% to 62.6%, and the low end is below the 52.4% you need to break even at -110. The sample size lesson shows why.
So even a real winning record goes into Kelly as a discounted number, never at face value.
Where should your Kelly probability come from?
From a price you made before you looked at the bet, not from a feeling about the game.
There are two honest sources:
- Your own model. A spreadsheet that projects the game and turns the projection into a probability. This is what Kelly was built for.
- A sharp market price with the margin removed. If a sharp sportsbook has the game at a price that implies 54% after you remove its margin, and a softer book offers you a better price, the gap is your edge. Devigging Explained shows how to remove the margin step by step.
What does not work is guessing. "I am pretty sure about this one" is not 70%, and Kelly will happily turn that guess into a stake big enough to hurt you. If you cannot write the probability down before you see the price, you do not have an input for Kelly. The expected value lesson shows how probability and price combine into an edge.
What fraction of Kelly should you use?
Start with a quarter. It is the fraction I would give any bettor whose numbers come from a simple model, which describes almost everyone.
A few rules make it work in practice:
- Calculate full Kelly, then multiply by 0.25. In the example above, 5.5% becomes about 1.4%.
- Do not bet when Kelly is zero or negative. That is the formula telling you there is no edge at this price.
- Recalculate from your current bankroll, at least once a week, so stakes shrink during a losing run.
- Move to half Kelly only after a long tracked record shows your probabilities are well calibrated and you beat the closing line.
If quarter Kelly feels too small, that is usually a sign the bankroll is too small, not the fraction. The Kelly Criterion lesson lets you try different edges, odds and fractions and see the stake change.
FAQ
Is the Kelly Criterion good for sports betting?
Yes, as long as you use a fraction of it and have a real probability to feed it. Kelly is the only common staking method that sizes each bet by both your edge and the odds, and it stakes nothing when there is no edge. Flat staking ignores both. The weak point is the input: Kelly assumes your probability is exact, and in sports betting it never is. Full Kelly on a slightly wrong number can lose money even when you have an edge. Quarter Kelly keeps most of the benefit and removes most of that risk, which is why it is the version I would use.
What is half Kelly and quarter Kelly?
Half Kelly and quarter Kelly mean staking one half or one quarter of the full Kelly stake. If full Kelly says 5.5% of your bankroll, half Kelly is 2.75% and quarter Kelly is about 1.4%. You calculate the full number first, then multiply by 0.5 or 0.25. Smaller fractions grow the bankroll more slowly, but they cut the swings by much more than they cut the growth. In a simulation of 1,000 bets with a real 5% edge, quarter Kelly still grew the bankroll by about 80%, and the chance of it ever falling to half was under 1%.
What does a negative Kelly number mean?
A negative Kelly number means you have no edge at this price, so you do not bet. It happens when your probability times the decimal odds is less than 1. For example, if you think a team wins 50% of the time and the price is -110 (1.91), the formula gives about -5%. In theory a negative number says to bet the other side, but only if your model also has a real number for that side at that side's price. In practice, treat any result of zero or below as a pass. The MAX(0, ...) part of the Google Sheets formula does this for you.
How do you use Kelly when you bet several games at once?
Stake a smaller fraction when you place several bets at the same time, because Kelly assumes one bet is settled before the next one starts. If you place eight bets on a Sunday at full Kelly each, your total stake can easily pass 30% of your bankroll at once, which is far more risk than the formula intended. The simple fix is to use quarter Kelly and cap the total of all open bets at a share of your bankroll you are comfortable with. Also be careful with bets that are linked, like a side and a total in the same game, since they tend to win or lose together.
Should you recalculate your bankroll after every bet?
Recalculating after every bet is the exact version of Kelly, but once a day or once a week is close enough for most bettors. The point is that stakes follow your current bankroll. After a winning run your stakes grow, and after a losing run they shrink, which is what protects you from going broke. If you never update the bankroll figure, you are flat staking with a Kelly label on it. Pick one fixed time, update the number in your sheet, and let every stake that week come from it.
Can you use the Kelly Criterion on parlays?
You can, but most parlays have no edge to size. Kelly works on any single price: put in your probability for the whole parlay and its decimal odds. The problem is that the bookmaker's margin is charged on every leg, so a parlay of fair-looking legs usually ends up with a negative edge, and Kelly returns zero. If you do have a real edge on each leg, put the parlay through the same formula and expect a small stake, because the chance of it winning is low.
If you want a number of your own to put into Kelly, my models for the NFL, NBA, NHL, MLB and soccer are on the models page. If you want the same price-first logic applied to a single game for you, that is what the AI Betting Assistant does.
Where do you start if you want a staking plan you can follow?
Kelly answers how much to stake once you have a probability. If you want to build that probability yourself and pair it with a simple staking plan, that is what The Ultimate Modern Bettor's Toolkit is for.
$27, one payment. It includes the +EV betting model spreadsheet, a 45-minute video training that builds your first model with you, and the four-bucket staking system. 30-day money-back guarantee.
MB built Underdog Chance around one idea: make your own number first, then bet only when the price beats it. Over the last 20+ years his simple, probability-based Google Sheets models have helped thousands of bettors stop following picks and start pricing bets for themselves.
Last updated: 30 October 2026