The Kelly Criterion explained: optimal bet sizing for gambling and investing
Last updated: April 2026
The Kelly Criterion is a mathematical formula that tells you exactly how much to bet when you have an edge. Developed by John Kelly at Bell Labs in 1956 for signal noise problems, it was quickly adopted by gamblers and investors because it solves a fundamental problem: if you have a profitable opportunity, how much should you risk to maximise long-term growth without going bust? The Kelly Criterion formula is elegant, powerful, and widely misused. Whether you call it the Kelly Criterion betting strategy or simply the Kelly method, the core idea is the same. This guide explains how it works, walks through real examples in horse racing and football, covers fractional Kelly for practical use, and is honest about when and why most people should use it cautiously.
Try the Kelly Criterion Calculator → Work out your returns instantly with our free kelly criterion calculator.
The Kelly Criterion formula
Kelly % = (bp − q) / b
Where:
- b = decimal odds − 1 (the net profit per £1 staked if you win)
- p = your estimated probability of winning
- q = 1 − p (probability of losing)
The result is the percentage of your bankroll you should stake.
Worked example — horse racing
You find a horse at 5.00 decimal odds (4/1) and estimate the true probability of winning at 25%.
b = 5.00 − 1 = 4.00 p = 0.25 q = 0.75 Kelly % = (4.00 × 0.25 − 0.75) / 4.00 = (1.00 − 0.75) / 4.00 = 0.0625 = 6.25%
Kelly says stake 6.25% of your bankroll. On a £1,000 bankroll, that's £62.50. This is the Kelly Criterion horse racing application in its simplest form — one selection, one price, one probability estimate.
Worked example — football
You estimate Arsenal has a 55% chance of beating Chelsea. The bookmaker offers 2.10 (implied probability 47.6%).
b = 2.10 − 1 = 1.10 p = 0.55 q = 0.45 Kelly % = (1.10 × 0.55 − 0.45) / 1.10 = (0.605 − 0.45) / 1.10 = 0.141 = 14.1%
Kelly says stake 14.1% of your bankroll. On a £500 bankroll, that's £70.50. Already this feels aggressive — and it is. This is the Kelly Criterion football betting answer, and it highlights why full Kelly is dangerous in practice.
Our Kelly Criterion calculator does these calculations instantly — enter the odds and your estimated probability and it shows the optimal stake plus half-Kelly and quarter-Kelly alternatives.
Why Kelly maximises growth
The Kelly Criterion is mathematically proven to maximise the long-term growth rate of your bankroll. No other staking strategy grows your bankroll faster, on average, over many bets.
It achieves this by balancing two competing risks:
- Staking too little wastes your edge — you grow slowly despite having an advantage
- Staking too much exposes you to ruin — a losing streak can wipe you out before your edge has time to play out
Kelly finds the exact point where growth is maximised and the risk of ruin is theoretically zero (over infinite bets). It's the mathematical sweet spot.
But "theoretically" is doing a lot of work in that sentence. In practice, Kelly has serious problems.
The problem with full Kelly
In theory, full Kelly is optimal. In practice, it's too aggressive for almost everyone.
The volatility problem
Full Kelly produces enormous bankroll swings. A 14% stake means your bankroll fluctuates wildly between bets. A run of 5 losses at 14% reduces your bankroll by 52%. That's gut-wrenching even if you know the maths says you'll recover.
| Consecutive Losses | Bankroll at 14% Kelly | Bankroll at 3% Kelly |
|---|---|---|
| 1 | 86.0% | 97.0% |
| 3 | 63.6% | 91.3% |
| 5 | 47.0% | 85.9% |
| 10 | 22.1% | 73.7% |
| 15 | 10.4% | 63.3% |
At full Kelly, 10 consecutive losses cuts your bankroll by nearly 80%. At 3% Kelly (roughly one-fifth of full), the same streak costs less than 27%. Both represent the same edge — but the experience of living through them is completely different.
The probability of 10 consecutive losses depends on your win rate, but even with a 55% win rate (strong for sports betting), a 10-loss streak has roughly a 0.03% chance — which sounds tiny until you consider that over thousands of bets, it's virtually certain to happen at some point. Our risk of ruin calculator shows the probability of losing your entire bankroll at any given staking level.
The estimation problem
Kelly is only as good as your probability estimate. If you think you have a 55% chance but the true probability is 48%, Kelly tells you to make a large stake on a bet that is actually −EV. You'll lose money faster than flat staking.
This is the critical issue: the people most confident in their probability estimates are often the most overconfident. Overestimating your edge by even 5 percentage points turns Kelly from an optimal strategy into a destructive one. The psychology of gambling shows that overconfidence is one of the most persistent cognitive biases in betting — we systematically believe we know more than we do.
Fractional Kelly — the practical solution
The fractional Kelly criterion is what most professional bettors and investors actually use — a fraction of the full Kelly stake:
| Approach | Stake | Growth Rate (vs Full) | Volatility |
|---|---|---|---|
| Full Kelly | 100% of Kelly % | 100% | Very high |
| Half Kelly | 50% of Kelly % | ~75% | Moderate |
| Quarter Kelly | 25% of Kelly % | ~50% | Low |
| Eighth Kelly | 12.5% of Kelly % | ~25% | Very low |
Half Kelly — the standard recommendation
You sacrifice roughly 25% of the theoretical growth rate but dramatically reduce volatility and the impact of probability estimation errors. In the Arsenal example above, half Kelly gives a 7.05% stake instead of 14.1% — still aggressive, but survivable.
Half Kelly is the most common choice among professional bettors and quantitative investors. It's the default recommendation unless you have exceptional confidence in your probability model.
Quarter Kelly — for cautious bettors
Appropriate when you're less confident in your estimates, working with a smaller bankroll, or in the early stages of testing a strategy. You sacrifice half the growth rate but gain substantial protection against estimation errors and bad runs.
When Kelly says don't bet
If the Kelly percentage comes out at zero or negative, it means you don't have an edge — the odds don't offer value given your probability estimate.
Example: bet at 2.00 decimal, your probability estimate 45% Kelly % = (1.00 × 0.45 − 0.55) / 1.00 = −0.10
Negative Kelly = don't bet. The mathematically correct action is to walk away.
This is one of Kelly's most valuable outputs. It forces honesty: if you can't identify a genuine edge, the formula tells you to stake nothing. Most recreational gamblers would benefit enormously from this discipline alone — the willingness to say "there's no bet here today."
Kelly for different sports
Kelly Criterion horse racing
Horse racing is where Kelly is most naturally applied. Individual races, clear odds, quantifiable form data. The typical Kelly application: analyse the field, estimate each horse's win probability, compare to the bookmaker's odds, and stake according to Kelly on any +EV selections.
The challenge: estimating win probabilities in horse racing is extremely difficult. A 20-runner handicap at Cheltenham involves dozens of variables — form, going, draw, weight, jockey, trainer, pace. Most bettors overestimate their ability to price these fields accurately.
Kelly Criterion football betting
Football presents different challenges. Three-way markets (home/draw/away) mean your probability estimates must account for all three outcomes summing to 100%. The bookmaker's margin is typically lower on football than racing (4-6% vs 10-20%), which means the available edge is smaller — requiring more precise probability estimates.
Kelly Criterion sports betting — general principles
Across all sports, Kelly works best when:
- You have a repeatable method for estimating probabilities
- Your sample size is large enough to validate your edge (thousands of bets)
- You're disciplined enough to use fractional Kelly
- You track results rigorously and adjust when the data shows your edge has shrunk
Kelly for casino games — the uncomfortable truth
Kelly says don't bet on any standard casino game. The Kelly Criterion blackjack application is the one partial exception — card counters can achieve +EV and use Kelly to size their bets accordingly, but casinos actively prevent this. Every casino game has a negative expected value for the player. The Kelly formula returns a negative percentage for every roulette bet, every slot spin, every baccarat hand. The mathematically optimal casino stake, according to Kelly, is £0.
This doesn't mean you shouldn't play casino games — it means the Kelly Criterion isn't the right framework for them. Casino gambling is entertainment with a known cost (the house edge). Kelly is designed for situations where you have a genuine edge, which casino games don't offer.
Kelly and bankroll management
The Kelly Criterion is a staking method, not a complete bankroll management system. It answers "how much to bet" but not "how much to set aside for gambling" or "when to stop."
A good staking plan built around Kelly looks like this:
- Set a gambling bankroll based on what you can afford to lose entirely
- Determine your edge through analysis and honest probability estimation
- Calculate the Kelly stake using the formula
- Apply fractional Kelly (half or quarter) for safety
- Track results over hundreds of bets to validate whether your edge is real
If at any point the numbers show you don't have an edge — or if you find yourself increasing stakes to chase losses rather than following the formula — that's a signal to step back. Our blog post 5 numbers every gambler should check covers the specific metrics worth tracking.
The Kelly Criterion graph — visualising the edge
If you plot bankroll growth rate against stake size, the Kelly Criterion sits at the peak of the curve. Staking below Kelly grows your bankroll slower than optimal. Staking above Kelly actually reduces your growth rate — and at 2× Kelly, your expected growth rate drops to zero. Beyond 2× Kelly, your expected bankroll shrinks over time despite having a positive edge.
This is counterintuitive: betting more than Kelly on a +EV bet actually makes you poorer in the long run. The volatility overwhelms the edge. It's a powerful argument for fractional Kelly — even if you're right about having an edge, overstaking destroys the advantage.
Common Kelly mistakes
Using Kelly on −EV bets. Kelly only works when you genuinely have an edge. Applying it to casino games or random sports bets without a validated probability model just means you're losing money in Kelly-sized chunks.
Trusting your probability estimate too much. The formula is precise; your inputs are not. A small error in probability estimation produces a large error in recommended stake. This is why fractional Kelly exists.
Ignoring correlated bets. Kelly assumes independent bets. If you're placing 5 football bets on the same Saturday, the outcomes may be correlated (weather, league trends). The total exposure can exceed what Kelly intends.
Using Kelly without tracking. If you're not recording every bet, every result, and reviewing your actual edge over hundreds of bets, you're not using Kelly — you're guessing stake sizes with a mathematical veneer. If you notice signs of problem gambling creeping in — staking more than the formula says, chasing losses, or gambling money you can't afford to lose — the Kelly framework has broken down and it's time to reassess.
Frequently asked questions
Written by
David Burke
David is a gambling industry analyst and poker player based between London, Spain, and Malta. He has spent over a decade observing the European betting and casino landscape, with particular expertise in odds, probability, game strategy, and how the bookmaking industry works. At WiseStaker, David writes guides on bet types, game rules, and the mathematics behind gambling.
Follow on X →