Bet the full Kelly amount on every edge you find and the math guarantees the fastest possible long-run growth of your bankroll. It also hands you a coin-flip’s chance — roughly 50% — of watching that bankroll get cut in half somewhere along the way. That gap, between what the formula promises and what it puts you through, is the whole story of Kelly Criterion betting. Used well, it is the sharpest staking tool in gambling; used carelessly, it is one of the fastest routes to ruin.
KEY FACTS AT A GLANCE
- The formula: f* = (bp − q) / b — plainly, your edge divided by the odds.
- What it does: Kelly does not find winning bets; it sizes bets you have already confirmed are positive-value so variance never wipes you out.
- Where the edge comes from: strip the vig out of the market price, then compare that fair number to your own probability. No gap, no bet.
- Full Kelly: the fastest growth, but a ~50% chance of your bankroll ever being halved along the way.
- Half Kelly: keeps about 75% of the growth for only ~25% of the volatility — which is why most pros bet a fraction.
- The real danger: overestimating your own win probability. Overbetting is punished far more harshly than underbetting.
ON THIS PAGE
What the Kelly Criterion Actually Is
The Kelly Criterion is a staking rule: given a bet you believe has an edge, it tells you exactly what fraction of your bankroll to wager to grow your money as fast as mathematically possible over the long run. It is named after John L. Kelly Jr., the Bell Labs physicist who derived it in 1956, and it has been used to beat casinos and markets ever since.
Here is the part most guides bury, so we will lead with it: Kelly does not make you a winning bettor. It cannot manufacture an edge you do not have. If you feed it bets with no value, it correctly tells you to bet nothing. Finding positive expected-value opportunities is what makes you profitable — Kelly’s only job is to size those opportunities so a bad run of variance never bankrupts you before your edge pays off. Think of it as the difference between the engine and the suspension: your edge is the engine, Kelly is what keeps you on the road.
That framing matters because Kelly is unlike doubling systems such as the Martingale, which try to conjure profit out of a game with no edge and inevitably fail. Kelly assumes you have done the hard work of finding value first — so before you size a single bet, confirm it is genuinely positive expected value.
Related tools: Kelly Criterion Calculator · No-Vig Calculator · Expected Value Calculator · Bankroll Calculator · Odds Converter
The Kelly Formula, Decoded
The betting version of the Kelly formula is short:
Each symbol is simple once you break it down:
| Symbol | Meaning |
|---|---|
| f* | The optimal fraction of your current bankroll to stake on this bet. |
| b | The net decimal odds — how much you win per unit staked. It equals the decimal odds minus 1 (decimal 2.50 → b = 1.5). |
| p | Your estimated probability of winning the bet (as a decimal, e.g. 0.55 for 55%). |
| q | Your probability of losing, which is simply 1 − p. |
The numerator, bp − q, is your edge: your expected net profit per unit staked. Divide that edge by the odds b and you get the slice of your bankroll to wager. This is why practitioners often quote Kelly as simply “edge divided by odds.” The rule falls out naturally: a bigger edge means a bigger bet, while longer odds (which carry more variance per unit) pull the stake back down. And if your edge is zero or negative, f* comes out at zero or below — Kelly’s built-in discipline telling you to pass.
The version that is faster to use
Because b is always the decimal odds minus 1, you can rewrite Kelly so it takes the decimal price straight off the screen with no subtraction step:
The top of that fraction, p × d − 1, is your expected return per dollar staked, which is the same number an expected value calculator gives you. So the working rule is even simpler than it looks: Kelly stake = expected value per dollar ÷ the profit per dollar. A bet returning 8 cents on the dollar at odds that pay $1.40 profit per dollar gets a 0.08 ÷ 1.40 = 5.7% stake.
There is one shortcut worth memorizing. At even money (decimal 2.00, b = 1), the formula collapses to f* = 2p − 1, which is exactly your edge. A 4% edge on a coin-flip-priced bet means a 4% stake — no arithmetic required. That is why card counters, whose blackjack edges run about 0.5% to 1.5% on close-to-even-money hands, end up betting roughly 0.5% to 1.5% of their bankroll per hand.
The formula needs decimal odds to work, so convert first. American +150 becomes decimal 2.50 (b = 1.5); American −110 becomes decimal 1.909 (b ≈ 0.909); a fractional quote like 6/4 is your b directly (1.5). If juggling formats trips you up, our odds converter handles it instantly, and our guide on how to read betting odds walks through each format.
One nuance worth knowing: this simple form is exactly right for betting, because a losing bet forfeits 100% of your stake. It should not be copied straight onto stock positions, where a “loss” is usually a partial drawdown rather than a total wipeout — that case needs a different version of the equation. For sports and casino wagers, though, f* = (bp − q) / b is the correct tool.
Worked Examples: From Edge to Dollars
Numbers make it click. Take a +150 underdog (decimal 2.50, so b = 1.5) that you have assessed as a genuine 55% chance to win. Your edge is bp − q = (1.5 × 0.55) − 0.45 = 0.375, and f* = 0.375 ÷ 1.5 = 0.25. Kelly says stake 25% of your bankroll — $250 on a $1,000 roll. Now suppose you only rate that same bet a true coin flip at −110 (decimal 1.909). The edge turns negative (−0.05), f* drops below zero, and Kelly tells you to bet nothing: the bookmaker’s vig means a 50/50 read is a losing proposition. The table below runs several bets end to end.
| The bet | Decimal | b | Your p | Edge (bp − q) | Kelly f* | Stake on $1,000 |
|---|---|---|---|---|---|---|
| Even money, slim edge | 2.00 | 1.0 | 53% | 0.06 | 6% | $60 |
| Even money, strong edge | 2.00 | 1.0 | 60% | 0.20 | 20% | $200 |
| +150 underdog | 2.50 | 1.5 | 55% | 0.375 | 25% | $250 |
| 2-to-1 payout coin | 3.00 | 2.0 | 50% | 0.50 | 25% | $250 |
| +140 dog, realistic edge | 2.40 | 1.4 | 45% | 0.08 | 5.7% | $57 |
| −110 favorite, no edge | 1.909 | 0.909 | 50% | −0.05 | 0% — no bet | $0 |
Notice how the stake tracks the edge, not the odds: the slim 53% even-money edge earns a cautious 6% bet, while the fat edges push toward a quarter of the roll. Also notice which row looks most like real life. The 55% and 60% reads are textbook illustrations; the +140 underdog you rate at 45% — an 8-cent edge and a 5.7% stake — is what an actual profitable bet looks like. Edges that justify staking a quarter of your bankroll are vanishingly rare outside of advantage play. You can run your own scenarios through our Kelly Criterion Calculator instead of doing the arithmetic by hand.
Finding the Edge: Where Your Probability Actually Comes From
Every Kelly guide hands you the formula and then quietly leaves the hardest input — p — as an exercise for the reader. That is backwards, because p is the only number in the equation you can get wrong. The odds are printed on the screen; your win probability is a claim you are making about the world. Here is how to make that claim defensible.
Step one: strip the vig out of the market price
Sportsbook odds are not probabilities. They are probabilities plus a built-in margin, which is why the implied chances of every outcome in a market add up to more than 100%. Take the standard −110 both ways: decimal 1.909 implies 1 ÷ 1.909 = 52.38% for each side. Add them and you get 104.76%. That extra 4.76% is the vig, and it belongs to the book, not to reality.
Removing it is a division problem. Divide each side’s implied probability by the total, and the market’s honest opinion falls out — for the −110/−110 market, 52.38 ÷ 104.76 = exactly 50% per side. Do the same on a lopsided game and it looks like this:
| Side | Price | Decimal | Raw implied | Fair (no-vig) |
|---|---|---|---|---|
| Home | −160 | 1.625 | 61.54% | 59.63% |
| Away | +140 | 2.40 | 41.67% | 40.37% |
| Total | — | — | 103.21% (3.21% vig) | 100% |
Our no-vig calculator does this in one click, and the vig calculator tells you how heavily a book is taxing a given market. One caveat for the purists: this proportional method is the standard quick approach, but it slightly overstates the fair chance of heavy favorites. On lopsided prices, treat the output as a very good estimate rather than gospel.
Step two: beat that number with your own
The no-vig figure is the market’s fair price, and the market is a formidable opponent — sharp books employ modelers, absorb millions in informed money, and move on news faster than you will. So the question is not “do I like this team?” It is: do I have a specific reason to think 40.37% is wrong? An injury the line has not digested, a lineup change, a weather read, a number your model produces that the market has not seen — something concrete.
Say you land on 45% for that +140 underdog. The market’s fair number is 40.37%, so you are claiming a 4.6-point disagreement. Run it through Kelly: f* = (1.40 × 0.45 − 0.55) ÷ 1.40 = 0.08 ÷ 1.40 = 5.71%. On the same $1,000 bankroll used above that is a $57 full-Kelly bet — or $29 at half Kelly, which is where you should actually be. That single number is the whole workflow: devig, disagree, size.
PRO TIP: LET THE CLOSING LINE GRADE YOUR ESTIMATES
You cannot audit p directly — the game only happens once. But you can audit it indirectly. Log the price you took and the no-vig closing price for every bet. If you are consistently getting better numbers than the close, your probability estimates contain real signal and Kelly is safe to run. If you are not beating the close, the honest conclusion is that your edge is imaginary, and no staking formula will save a negative expectation.
How to Apply the Kelly Criterion, Step by Step
Put together, the whole method is seven moves. Once you have done it a dozen times it takes under a minute per bet.
- Define your bankroll. Fix a single number that is money you can afford to lose, kept separate from everyday funds. Every Kelly stake is a percentage of this figure, so it has to be real and it has to be one number.
- Devig the market to find the fair price. Convert every outcome to implied probability, add them up, and divide each by the total. That gives you the market’s honest opinion with the bookmaker’s margin removed.
- Set your own probability, p. Only override the market when you have a concrete reason — an injury, a lineup change, a model output. If your number matches the fair price, there is no bet here.
- Convert the odds and compute f*. Turn the price into decimal odds, subtract 1 to get b, then run f* = (bp − q) / b. A zero or negative result means pass, and passing is a result.
- Multiply by your Kelly fraction. Take a quarter to a half of f*, never more. This is not caution for its own sake — it is the buffer that absorbs the error in step three.
- Apply your cap, then place the bet. Multiply the fractional f* by your bankroll to get the dollar stake, and cut it back if it breaches your maximum single-bet limit. A stake that looks enormous is usually a sign that p is wrong, not that you found a gift.
- Update the bankroll and repeat. Recalculate from your new balance before the next bet. This automatic scaling — smaller stakes after losses, larger after wins — is the mechanism that keeps you solvent.
Full Kelly vs Fractional Kelly
Those raw f* figures are full Kelly — the growth-maximizing stake. In practice almost no serious bettor uses them at face value. Instead they bet a set fraction of Kelly: “half Kelly” (multiply f* by 0.5), “quarter Kelly” (× 0.25), and so on. To see why, it helps to look at what happens to your long-run growth as you scale your bet size up and down around the full-Kelly point.
THE KELLY CURVE
Growth follows a hill. It climbs to a single peak at full Kelly, then falls away — and because the curve is flat near the top, betting only half Kelly still captures about three-quarters of the maximum growth. Underbetting costs you a little; overbetting costs you everything.
The trade-off is lopsided in your favor. Cutting your stake in half surrenders only about a quarter of your growth rate, but it slashes your volatility to roughly a quarter of full Kelly's. In exchange for giving up a little compounding speed, you buy a dramatically smoother, safer ride — the exact trade most bankroll-conscious bettors want. The chart below lays the two effects side by side.
This is why most professionals live in the quarter-to-half-Kelly range. Researchers who have modeled famous investors put Warren Buffett close to a full-Kelly bettor and John Maynard Keynes at around 80% — but those are people with deep edges and iron stomachs. For a sports bettor whose edge is an estimate, half or quarter Kelly is the sane default. Whatever fraction you choose, size it against a defined bankroll; our bankroll calculator helps you set that number before you start.
What a losing streak actually does to your stakes
The single most useful property of any Kelly-style plan is invisible until you hit a bad run. Because every stake is a percentage of your current bankroll, your bets shrink automatically as you lose. You can approach zero, but you can never mathematically reach it — there is always something left to bet with.
Watch what that looks like against flat betting. Start with $1,000 and a bet you rate 55% at +150, so half Kelly is 12.5% of the roll. A flat bettor sizing the same first wager stakes $125 every time. Then eight straight losses arrive — which sounds apocalyptic but is a 1-in-595 sequence at 55%, meaning it will happen to anyone who bets a full season.
| Loss # | Half-Kelly stake | Bankroll after | Flat $125 bankroll |
|---|---|---|---|
| 1 | $125.00 | $875.00 | $875.00 |
| 2 | $109.38 | $765.63 | $750.00 |
| 3 | $95.70 | $669.92 | $625.00 |
| 4 | $83.74 | $586.18 | $500.00 |
| 5 | $73.27 | $512.91 | $375.00 |
| 6 | $64.11 | $448.80 | $250.00 |
| 7 | $56.10 | $392.70 | $125.00 |
| 8 | $49.09 | $343.61 — still betting | $0 — busted |
The flat bettor is finished on bet eight. The Kelly bettor still has $343.61 and an edge, which is the entire point: surviving the streak is what lets the edge eventually pay. The flip side is that percentage staking recovers more slowly, since you are climbing back with smaller bets. That is the price of never going broke, and it is a bargain.
The Overbetting Trap
Look again at the right-hand side of the Kelly curve. Push past full Kelly and growth does not just slow — it collapses. Bet exactly double the Kelly fraction and your long-run growth rate falls all the way to zero; go beyond that and it turns negative, marching you toward certain ruin no matter how real your edge is. This is not a fringe result: it is the same mistake, at institutional scale, that helped blow up the hedge fund Long-Term Capital Management.
"The growth rate becomes zero plus the risk free rate when one bets exactly twice the Kelly wager. Hence it never pays to bet more than the Kelly strategy."
— MacLean, Thorp & Ziemba, "Good and Bad Properties of the Kelly Criterion"
Even betting the correct full Kelly amount is a white-knuckle experience. Because the stake is a fixed fraction of a bankroll that keeps shrinking on losing runs, full Kelly carries a roughly 50% chance that your bankroll will, at some point, be cut in half. Scale back to half Kelly and that chance falls to about 12.5%; at quarter Kelly it is under 1%. The chart makes the safety gap stark.
What happens when your probability is wrong
Here is the trap most bettors fall into without realizing it. The Kelly formula assumes you know your true win probability p. You almost never do — you estimate it. And the penalty for guessing too high is not proportional; it compounds, because an inflated p raises your stake at the same time it lowers the correct stake. You move away from the target from both directions at once.
Take the article's headline bet: +150, decimal 2.50, and you have rated it 55%, so you confidently stake 25% of your bankroll. Now suppose reality disagrees. The table shows the Kelly stake that would have been correct at each true probability, and where your 25% bet actually lands on the curve.
| True p | Correct Kelly stake | Your 25% bet = | Half Kelly (12.5%) = | Verdict on the 25% bet |
|---|---|---|---|---|
| 58% (you were modest) | 30.0% | 0.83× Kelly | 0.42× Kelly | Fine — 97% of max growth |
| 55% (you were right) | 25.0% | 1.00× Kelly | 0.50× Kelly | Optimal — peak growth |
| 52% (off by 3) | 20.0% | 1.25× Kelly | 0.63× Kelly | Overbetting — 94% of growth |
| 50% (off by 5) | 16.7% | 1.50× Kelly | 0.75× Kelly | Overbetting — 75% of growth |
| 47% (off by 8) | 11.7% | 2.14× Kelly | 1.07× Kelly | Past the cliff — growth is negative |
| 45% (off by 10) | 8.3% | 3.00× Kelly | 1.50× Kelly | Bleeding out — ruin trajectory |
| 42% (off by 13) | 3.3% | 7.50× Kelly | 3.75× Kelly | Catastrophic on both |
Read the two middle columns against each other and the case for fractional Kelly stops being a matter of temperament. Betting full Kelly on your estimate, you cross into negative growth on an 8-point probability error — the kind of miss that is completely routine when you are handicapping games. Betting half Kelly, you are still in positive-growth territory at a 10-point error, and even at 1.50× you keep 75% of the available growth. The half-Kelly discount is not timidity; it is the insurance premium you pay on the one input you cannot verify.
The asymmetry is what gets people. Underestimating your edge costs you a slice of growth. Overestimating it can cost you the bankroll. And because bettors are systematically overconfident, most people who believe they are betting full Kelly are in fact somewhere to the right of it.
THE ONE RULE THAT MATTERS MOST
Never overbet. When you are unsure of your true edge — and you always are — bet a fraction of Kelly, not more. In one controlled study where players were handed a coin they were told landed heads 60% of the time, 28% still went broke by betting too aggressively. They knew the edge exactly, and a quarter of them still found a way to lose. The formula only protects the disciplined.
Kelly with Multiple and Correlated Bets
The formula solves for one bet at a time, resolved before the next one starts. Real betting is messier: you often have four tickets live on a Sunday afternoon, and some of them are quietly the same bet in different clothes. Three situations come up constantly.
| Situation | Why the plain formula misfires | What to do instead |
|---|---|---|
| Several unrelated games at once | Each stake is correct alone, but they are all at risk simultaneously, so your true exposure is the sum. | Compute each f* normally, then scale every stake down proportionally so the combined exposure fits a limit you set in advance. |
| A parlay or same-game combo | The legs are correlated, so multiplying their individual probabilities gives you the wrong p. | Treat the whole ticket as one bet. Estimate the joint probability of all legs landing, use the combined price as your odds, and apply Kelly once. |
| The same side at two books | Sizing each ticket at full Kelly doubles your position on a single opinion. | Size the position once, then split that dollar amount across the books. Two tickets, one Kelly stake. |
The parlay case is worth dwelling on, because it is where the correlation problem bites hardest. Kelly's derivation assumes independence, and a parlay is a machine for violating that assumption — a quarterback's passing yards and his team's total points do not move independently. Books price same-game combos with that correlation baked in and a fat margin on top. If you insist on building them, size the ticket from its own joint probability, and expect the honest answer to be a very small number. If you are spreading risk across several outcomes instead, our dutching calculator and hedge calculator handle the split.
PRO TIP: FOUR GUARDRAILS THAT KEEP KELLY HONEST
- Hard-cap every bet. Pick a ceiling — many bettors use 5% of bankroll — and never exceed it no matter what f* says. An enormous Kelly number is almost always a broken probability, not a gift.
- Set a minimum edge. Below roughly 2%, your calculated edge is smaller than the error in your own estimate. Those bets are noise dressed as value.
- Recalculate on a schedule. Update your bankroll figure at fixed points — after each bet, or once a week — rather than adjusting mid-slate on a hot streak.
- Never re-size to chase. Kelly already tells you to bet less after losses. Overriding it to win the money back is the exact behavior the formula exists to prevent.
Kelly vs Other Staking Plans
Kelly is one option among several, and it is not automatically the right one. What separates the plans is how much they ask you to know, and how badly they punish you for being wrong.
| Staking plan | How stakes move | Can it bust you? | Best for |
|---|---|---|---|
| Flat / unit betting | The same dollar amount every time, regardless of edge. | Yes — a long enough streak reaches zero. | Beginners, and anyone whose probability estimates are shaky. |
| Fixed percentage | A set share of the current bankroll (say 2%), ignoring edge and price. | No — it can only approach zero. | Simple discipline without any modeling work. |
| Full Kelly | Edge ÷ odds, scaling with both your advantage and the price. | Not literally — but a 50% drawdown is a coin flip. | Proven, measurable edges such as advantage play. |
| Fractional Kelly (¼–½) | A set fraction of the Kelly stake, absorbing estimation error. | No — and drawdowns are far shallower. | Almost everyone. The working default. |
| Martingale & progressions | Stakes rise after losses, chasing the deficit back. | Yes — ruin is close to guaranteed. | Nothing. It cannot turn a losing bet into a winning one. |
The dividing line runs between the top two rows and the Kelly rows: flat and fixed-percentage staking need no opinion about your edge, while Kelly demands a number and rewards or punishes you according to how good that number is. If you are not yet confident your probabilities beat the closing line, flat betting is genuinely the better plan — our unit calculator will set your unit size. The bottom row is a different species entirely: progressions like the Martingale move in exactly the opposite direction to Kelly, staking more as your bankroll shrinks.
Where the Kelly Criterion Came From
The formula was born not at a racetrack but inside Bell Labs. In 1956, physicist John L. Kelly Jr. published a paper showing that a gambler with an information advantage could grow their bankroll exponentially at a rate equal to the rate of information itself — a direct application of Claude Shannon's information theory. Shannon, by then a colleague, encouraged him to publish. Kelly, who many rated the smartest person at Bell Labs after Shannon himself, died young at 41, but his staking rule outlived him.
It reached the gambling world through mathematician Edward O. Thorp, who learned of Kelly's paper from Shannon around 1960 and immediately put it to work. Thorp used Kelly-sized bets to beat blackjack through card counting — the subject of his landmark 1962 book Beat the Dealer — and later ran Princeton Newport Partners, a market-neutral hedge fund that compounded at roughly 19% a year for around two decades without a losing year. From a Bell Labs journal to blackjack tables to Wall Street, the same one-line formula did the work.
The Limits of Kelly (and How to Stay Honest)
Kelly is powerful, but it is not gospel. The economist Paul Samuelson spent years arguing against treating it as the universally "rational" staking rule — famously making his case in a paper written almost entirely in one-syllable words. His point still stands: maximizing the long-run growth of your money is only the right goal if your personal appetite for risk happens to match what the math assumes. Anyone more cautious than that should bet less than Kelly, not exactly Kelly. Beyond that, the formula leans on assumptions the real world routinely breaks.
WHEN KELLY HELPS — AND WHEN IT HURTS
Where Kelly earns its keep
- You have a genuine, measurable edge (advantage play, confirmed +EV bets)
- You bet repeatedly and reinvest from a single, growing bankroll
- You use a fraction (¼–½) to absorb the fact that your edge is an estimate
- You size by the math, never by gut feel or chasing losses
Where Kelly gets you hurt
- Your "edge" is really a guess — an 8-point miss puts you past the cliff
- Bets are correlated, as in a parlay or the same game across books
- One-off or short-run wagers, where the long-run guarantee never arrives
- Betting limits stop you scaling, so the math and your account disagree
- You bet full Kelly or more and a routine 50% drawdown forces you out
Two of those deserve a flag of their own. The simple formula assumes each bet is independent, which is exactly what a parlay or a slate of related wagers is not. And it assumes you can always bet the size it recommends — but a book that has noticed you win will cut your limits long before your bankroll grows into them, which is the practical ceiling most winning bettors hit first. No staking formula, however elegant, changes the fact that betting carries risk. Kelly is a tool for managing a bankroll you can afford to lose — never a reason to bet money you cannot. If gambling stops being fun, our responsible gambling resources are there to help.
The Bottom Line on Kelly Criterion Betting
Strip away the Bell Labs mythology and Kelly is a single, unglamorous instruction: bet in proportion to your edge, divided by your odds, as a share of what you currently have. That is it. The formula is trivial arithmetic; everything difficult about using it lives upstream, in the probability you feed it. Devig the market, disagree with it for a reason you can name, and you have earned the right to use the equation. Skip that work and Kelly will faithfully compute a precise, confident, entirely fictional stake.
So use it, but use it at a discount. Quarter to half Kelly costs you a modest slice of theoretical growth and buys you a buffer wide enough to survive being wrong about the one thing you are guaranteed to be wrong about. The bettors who blow up are not the ones who bet too little — they are the ones who trusted their own numbers a bit too much and let a beautiful formula do exactly what they told it to.
KEY TAKEAWAYS
- Kelly sizes bets, it doesn't win them — the formula only works on wagers you have already confirmed are positive value.
- The math is edge ÷ odds — f* = (bp − q) / b, using decimal odds; a zero or negative answer means don't bet.
- Your edge starts with devigging — strip the margin out of the market price, then justify why your own number is different.
- Full Kelly is dangerous — it maximizes growth but carries a ~50% chance of ever halving your bankroll.
- Fractional Kelly is the real-world default — half Kelly keeps ~75% of the growth for ~25% of the volatility; most pros bet a quarter to a half.
- Overbetting is the cardinal sin — double Kelly zeroes your growth, and an 8-point error in your probability is enough to get you there.
FAQs
Not on its own. Kelly only sizes bets that already have a positive expected value — it cannot create an edge. Finding value bets is what makes you profitable; Kelly's job is to size them so variance keeps you solvent long enough for that edge to pay off.
It is f* = (bp − q) / b. Here f* is the fraction of your bankroll to stake, b is the net decimal odds (decimal odds minus 1), p is your estimated probability of winning, and q is 1 − p. The top of the fraction, bp − q, is your edge, so the rule is simply edge divided by odds.
Start from the market. Convert every outcome in the market to an implied probability, add them up, and divide each one by that total — this strips out the bookmaker's vig and leaves the market's fair price. Then only override that number when you have a concrete reason, such as an injury or a model output. If your probability matches the fair price, there is no edge and no bet.
Because the trade-off strongly favors it. Betting half the Kelly stake keeps roughly 75% of the long-run growth while cutting volatility to about a quarter, and it lowers the chance of ever halving your bankroll from around 50% to about 12.5%. Half Kelly also cushions the near-certain error in your probability estimate.
It depends on how accurate your probabilities are. Kelly maximizes long-term growth when your win estimates are reliable, but it is far less forgiving of mistakes. Flat betting a fixed unit is steadier and more forgiving of estimation error, which often makes it the better choice for less experienced bettors.
Your growth falls even though your stake rises. Betting double the Kelly fraction drops your long-run growth rate to zero, and anything beyond that turns it negative and trends toward ruin. Because most people overestimate their edge, accidental overbetting is the most common and most dangerous Kelly mistake.
Yes — it is one of the most widely used bankroll-management methods in sports betting. The catch is that your edge is always an estimate, and sports markets are high-variance, so most bettors apply fractional Kelly (a quarter to a half) rather than the full stake to stay on the safe side.
Treat a parlay as a single bet, because its legs are correlated and multiplying their individual probabilities gives the wrong answer. Estimate the joint probability of every leg landing, use the combined price as your odds, and run Kelly once. For several unrelated bets live at the same time, calculate each stake normally and then scale them all down proportionally so your total exposure stays inside a limit you set in advance.
Only where you actually have an edge, such as advantage play or a confirmed positive-value bet. Against a normal house edge, the formula returns zero or a negative number — its way of telling you to bet nothing. On games where the house holds the advantage, Kelly simply confirms there is no stake worth making.
Far more than most bettors expect, because an inflated probability raises your stake at the same time it lowers the correct stake. Rate a bet at 55% when it is really 47% and the stake you thought was full Kelly is actually 2.14 times the right size — past the point where long-run growth turns negative. At half Kelly the same error leaves you at 1.07 times, still safely in positive territory, which is the whole argument for betting a fraction.
Sources
- A New Interpretation of Information Rate (1956) — John L. Kelly Jr., Bell System Technical Journal
- The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market (2006) — Edward O. Thorp
- Good and Bad Properties of the Kelly Criterion (2010) — MacLean, Thorp & Ziemba
- Rational Decision-Making Under Uncertainty: Observed Betting Patterns on a Biased Coin (2016) — Victor Haghani & Richard Dewey
- Why We Should Not Make Mean Log of Wealth Big Though Years to Act Are Long (1979) — Paul A. Samuelson
- The Kelly Criterion: You Don't Know the Half of It (2018) — CFA Institute
