Kelly criterion betting answers the question most bettors never ask out loud: how much should you actually stake? The formula turns your edge and your bankroll into one specific number. It doesn’t pick winners. What it does is stop a thin edge from ruining you, and stop you from overpricing a bad opinion.
Here’s the whole idea in one line. Stake the fraction of your bankroll that matches your edge relative to the price you’re getting. Everything below is detail.
What is Kelly criterion betting?
The Kelly formula returns the fraction of your bankroll to bet on a repeated wager where you have an edge. The classic version looks like this:
f = (bp – q) / b
Here f is your stake as a decimal (0.05 means 5% of bankroll). b is the net odds you’re getting. p is your estimated win probability. q is 1 – p.
Net odds are where people slip, so keep a shortcut handy. A price of +150 gives b = 1.5. A price of -110 gives b = 100/110, or 0.909. Plus prices convert straight to decimal. Minus prices need that divide.
The math comes from John Kelly at Bell Labs in 1956. He was studying how information moves through noisy phone lines, not football. The formula fit any repeated bet with an edge, which is why it’s still the standard. Wikipedia has the original derivation if you want the proof.
One warning before we go further. The formula has no opinion about whether your probability estimate is any good. That part is on you.
How do you calculate a Kelly stake?
Multiply your edge by the odds, then divide by the odds. Two examples make it concrete.
Say you think a bet wins 55% of the time and the price is +100. That gives b = 1, p = 0.55, and q = 0.45. So f = (1 x 0.55 – 0.45) / 1, which is 0.10. Kelly says 10% of your bankroll.
Now keep the same 55% and move the price to -110. With b = 0.909, f = (0.909 x 0.55 – 0.45) / 0.909, which is 0.055. Your stake drops to 5.5%.
Same opinion, half the bet. That’s Kelly’s most useful trait. It reacts to the price, not to how strongly you feel about the game.
Here’s what the formula spits out across a few common prices. Notice the 60% row at -150. Break-even on that price is exactly 60%, so Kelly says pass.
| Your win probability | +100 | -110 | -150 | +150 |
|---|---|---|---|---|
| 55% | 10.0% | 5.5% | No bet | 25.0% |
| 60% | 20.0% | 16.0% | 0% | 33.3% |
Now read the plus-money column again. At a 55% estimate, a +150 bet calls for 25% of your bankroll. That number should make you uncomfortable.
Why full Kelly is too aggressive for most bettors
Full Kelly assumes you know p exactly. You don’t. You have an estimate, and estimates move. When your number is off, full Kelly costs you more than it earns you, because the stake grows faster than your confidence deserves.
The standard fix is fractional Kelly. Bet a quarter or a half of what the formula says. You give up a little theoretical growth and take a lot less drawdown. Full Kelly math also implies deep swings along the way, and those only sting more when real money is on the line.
StatsBench builds that idea into its staking presets. Safe runs a 0.15 fraction of full Kelly, Balanced runs 0.30, and Risky runs 0.50, with a hard 5% cap on any single bet. That cap does more for your survival than the fraction does. No formula should ever put a quarter of your roll on one game.
A simple rule to live by: if full Kelly tells you to bet 20%, bet 5% and sleep fine. The Kelly criterion calculator will run the numbers for any price and probability you give it.
Where does your win probability come from?
Everything above depends on p, and p is the part nobody hands you. You build it from data you trust, then discount it for how thin the sample is.
Here’s a real line from the StatsBench cheatsheet on October 11, 2026. Brian Thomas Jr. receiving yards over 24.5 showed a 60% hit rate, 6 of 10 games, with a best available price of -111. Break-even at -111 is 52.6%, so the raw gap looks like 7.4 points.
Then look at the sample. Six of ten is ten games. Over a run that short, a 45% player and a 75% player can both land on 6 of 10. That isn’t a reason to skip the bet. It’s a reason to size it like the estimate it is. Our guide to sample size in betting covers how many games it takes before a hit rate means something.
Say your own work lands on 55% for that line instead of 60%. Full Kelly at -111 gives b = 0.901, so f = (0.901 x 0.55 – 0.45) / 0.901, about 0.05. Five percent of bankroll. Half Kelly would be 2.5%.
On a $2,000 bankroll that’s a $100 stake at full Kelly, or $50 at half. Most people would guess somewhere in that range and then fire $200 anyway. Writing the number down first is the whole point.
Getting to a fair probability takes two extra steps. Strip the book’s margin out of the market so you’re comparing against a real price, which is what the vig and hold calculator is for. Then remember that a hit rate and an edge are different numbers. Our breakdown of hit rate vs edge explains why a 70% prop can still be a losing bet at the wrong price.
Context matters too. A receiving line depends on target share, coverage, and game script, not just last month’s box scores. Our Trevor Lawrence props reader covers that same Jacksonville offense from the quarterback side, and the NFL props board puts every line in one sortable table.
Common Kelly mistakes
- Using a hot streak as your probability. Five good games is a mood, not an estimate.
- Running full Kelly on every bet. Even sharp bettors are wrong about p more often than they think.
- Ignoring the price. The same 55% estimate is a 10% bet at +100 and a 5.5% bet at -110. At -120, break-even alone is 54.5%.
- Shopping one book. Best price changes your break-even and your stake. StatsBench compares prices across roughly 60 sportsbooks so you’re not guessing.
- Treating a parlay as one Kelly bet. Correlated legs don’t multiply the way the formula assumes.
- Chasing after a loss. Kelly never tells you to bet more because you’re down. That’s tilt, and it’s the fastest route to a zero bankroll.
One more habit: keep a record. You can’t tell whether your estimates are any good without a log, and the Bet Tracker is free with any StatsBench account. It stores stake, price, and result, so you can grade your process instead of your memory.
Kelly vs flat betting
Flat betting means the same stake every time. It’s simpler, easier to stick to, and it never asks you to trust one probability estimate with your whole plan.
Kelly sizing scales with your edge and shrinks when the price is bad. When your estimates are decent, it grows a bankroll faster than flat betting. When your estimates are bad, it tells you so loudly.
A useful hybrid: keep flat units, then run the Kelly math as a sanity check. If the formula says 5% and you’re about to fire 8 units, one of those numbers is wrong, and it’s probably yours.
Variance is the other half of staking. A plan controls how much damage a losing run does, not whether one shows up. Our piece on variance in betting covers what a long sample looks like when you’re on the right side of the math.
Bet within your means, treat this as research and entertainment rather than income, and set limits before you start. If gambling stops being fun, 1-800-GAMBLER is a free call, and the National Council on Problem Gambling has more resources. 21+.
Free calculator
Run the numbers with the StatsBench Kelly criterion calculator
Instant, free, no account.
Open the Kelly criterion calculator →
Prefer email? Get the daily picks newsletter.
Frequently Asked Questions
What is the Kelly criterion in simple terms?
It’s a bet-sizing formula. You feed it your estimated win probability and the price you’re getting, and it returns the percentage of your bankroll to stake. Bigger edge means a bigger bet, worse price means a smaller one.
Should I use full Kelly or fractional Kelly?
Most bettors should use a fraction, usually a quarter to a half of what the formula says. Full Kelly assumes you know your true win probability exactly, and nobody does. Fractions cut the drawdowns while keeping most of the growth.
What win probability should I enter into a Kelly calculator?
Use your own honest estimate, not a hit rate taken at face value. A 6 of 10 sample is ten games, so shade it toward the middle before you size anything. If you can’t justify the number in one sentence, size down or pass.
Does the Kelly criterion guarantee a profit?
No. Kelly is a staking method, not a prediction. It only helps when your probability estimates are genuinely better than the market’s, and even then a losing run can last longer than you expect.
Can I use Kelly sizing for parlays?
Not directly. The formula assumes each bet stands alone, and parlay legs are often correlated, so the real probability isn’t the simple product of the legs. Size parlays smaller than the formula suggests, or treat them as entertainment.