The Kelly Criterion Explained: Optimal Staking for Football Betting
The Kelly Criterion is a mathematical formula that calculates the theoretically optimal stake for any bet when you know your edge.
It's powerful. But it requires honest probability estimates, which most bettors overestimate.
This guide covers the formula, worked examples, and practical application.
The Kelly Formula
f* = (bp - q) / b
Where:
- f* is the fraction of your bankroll to stake
- b is the odds minus one (decimal odds - 1)
- p is your probability of winning
- q is your probability of losing (1 - p)
Example: You think a team has 60% chance of winning. Odds are 2.0.
b = 2.0 - 1 = 1.0 p = 0.60 q = 0.40
f* = (1.0 * 0.60 - 0.40) / 1.0 = (0.60 - 0.40) / 1.0 = 0.20 / 1.0 = 0.20
Kelly says bet 20% of your bankroll.
Bankroll: 1000 pounds. Stake: 200 pounds.
Understanding the Output
Kelly outputs a percentage of your bankroll. This percentage is theoretically optimal for growth, assuming your probability estimates are accurate.
If Kelly says 5%, you bet 5% of your current bankroll on that match.
If Kelly says 25%, you bet 25% of your current bankroll.
Kelly assumes you're betting at the exact true probability. Small errors in probability dramatically change the output.
Why Most Bettors Can't Use Full Kelly
The formula assumes perfect probability estimates. In reality, you have edge.
Most bettors overestimate their edge. A system that's actually 52% accurate feels like 56%. This overestimation means Kelly output is too large.
Full Kelly based on overestimated edges leads to massive swings and eventual bust.
This is why fractional Kelly exists.
Kelly Criterion Worked Example
You've developed a model for football matches. It says:
Team A vs Team B: You estimate Team A at 55% to win. Odds are 2.4.
b = 2.4 - 1 = 1.4 p = 0.55 q = 0.45
f* = (1.4 * 0.55 - 0.45) / 1.4 = (0.77 - 0.45) / 1.4 = 0.32 / 1.4 = 0.229
Kelly says bet 22.9% of your bankroll.
Bankroll: 2000 pounds. Stake: 458 pounds.
This is a large stake because the odds (2.4) are good and your edge (55% vs 50% implied) is solid.
Why Kelly Works (In Theory)
Over many bets, Kelly maximises long-term bankroll growth.
With a genuine 55% win rate at 2.0 odds and proper Kelly stakes, your bankroll compounds faster than any other staking method.
But only if your probability estimates are accurate.
If you're overestimating (you're really 52% not 55%), Kelly leads to ruin.
The Danger of Overconfidence
A Kelly stake of 22% feels aggressive. It is.
If you're overestimating probabilities by 5%, that 22% stake is now sized for -0.5% edge. You're the underdog on a bet you think you're winning.
This is why Kelly fails for most bettors. They overestimate edges.
Kelly in Practice
Some professional bettors use Kelly. Most use fractional Kelly.
Fractional Kelly means using a fraction of the Kelly output. Half Kelly (0.5 * Kelly output) or Quarter Kelly (0.25 * Kelly output).
Half Kelly turns our 22.9% stake into 11.45%. Still respectable, but less aggressive.
Quarter Kelly turns it into 5.7%. Much safer.
Advantages of Kelly
- Theoretically optimal for long-term growth
- Automatically scales stakes with edge
- Accounts for odds and probability together
- Intellectually satisfying (it's based on math)
Disadvantages of Kelly
- Requires accurate probability estimates (most bettors overestimate)
- Stakes can be extremely aggressive
- One bad probability estimate ruins the calculation
- Bankroll swings are larger
When Kelly Makes Sense
Kelly works best if:
- You have a proven method with an edge over 100+ bets
- You track probabilities and verify your estimates are accurate
- You use fractional Kelly (1/4 or 1/2), not full Kelly
- Your bankroll is large enough to absorb variance
- You're okay with potentially large stakes
Calculating Kelly for Your Whole Portfolio
Some bettors calculate Kelly for individual bets. Others calculate the sum across a portfolio.
If you're placing five bets per week, Kelly might say 5%, 8%, 3%, 6%, 4%.
That's 26% total. Some bettors cap portfolio Kelly at 25%, so they'd reduce each stake proportionally.
This portfolio approach prevents overexposure from multiple Kelly bets in one week.
Kelly and Bankroll Size
Kelly stakes assume a large bankroll. A Kelly output of 22% of a 1000 pound bankroll is 220 pounds.
One bad bet and you're down 20%. With smaller bankrolls, this variance is dangerous.
Professional bettors use Kelly because they have 50000+ pound bankrolls. Swings are annoying but not threatening.
Moving From Kelly to Fractional Kelly
If you calculate full Kelly and the output feels too aggressive, use fractional Kelly.
Full Kelly: 20% Half Kelly: 10% Quarter Kelly: 5%
Quarter Kelly gives you most of the benefits (scaling with edge and odds) without the aggression. Recommended for most.
In Summary
- Kelly Criterion is mathematically optimal if your probability estimates are accurate.
- But most bettors overestimate their edge, making full Kelly dangerous.
- Use Kelly if you've proven your method over 200+ bets and are confident in probability estimates.
- Use fractional Kelly (quarter or half) for safety.
- Or skip Kelly entirely and use percentage or flat staking.
- Simpler methods work fine if executed consistently.
Frequently Asked Questions
Can I use Kelly if I've only placed 50 bets? Not reliably. You need at least 200 bets to understand your real win rate and edge. Using Kelly with limited data leads to overconfidence.
What's the difference between Kelly and Fractional Kelly? Full Kelly uses the complete output (e.g., 20% stake). Fractional Kelly uses a fraction (e.g., 1/4 Kelly is 5% stake). Fractional is safer and recommended for most bettors.
Should I use Kelly for all bet types? You can, but it's complex. Singles are easier. For accumulators, Kelly gets complicated because you're betting multiple events. Stick to flat or percentage staking for complex bets.
What if Kelly tells me to bet 0.5% of my bankroll? That's valid. It means the odds don't offer enough edge to warrant a large stake. Some bet types or markets offer less edge than others. A 0.5% stake is fine.
Can I adjust my Kelly stakes if they feel wrong? Yes. If Kelly says 20% but it feels too aggressive, use fractional Kelly instead. Trust your gut if the output seems unreasonable.
What if my probability estimate is 5% off? Then Kelly fails. That's the real risk. Which is why fractional Kelly is safer. It's less sensitive to probability estimation errors.