Kelly Criterion

What is the Kelly Criterion?

The Kelly Criterion is a mathematical formula that tells you exactly what percentage of your bankroll you should bet based on the edge you have. It was developed in 1956 by John Larry Kelly Jr., a researcher at Bell Labs, and originally applied to information theory. Decades later, the betting and finance world adopted it as the standard for optimal capital management.

The central idea is elegant: bet more when your edge is large, less when it’s small, and nothing when you have no edge. Sounds like common sense, but the formula gives you the exact number — not an approximation based on feelings.

What makes the Kelly Criterion special compared to other staking methods is that it maximizes your bankroll’s long-term growth. It doesn’t maximize the profits of a single bet or minimize the risk of a specific session. It optimizes compound growth, which is what truly matters if you think of betting as a long-term activity.

How does it work?

The formula is: f = (bp - q) / b

Where f is the fraction of your bankroll you should bet, b is the decimal odds minus 1, p is your estimated probability of winning, and q is the probability of losing (1 minus p).

Let’s put it in practice with a Premier League match. Manchester City plays at home against Brighton. Your analysis says City has a 60% probability of winning. The odds are 1.80.

b = 1.80 minus 1 = 0.80 p = 0.60 q = 0.40

f = (0.80 × 0.60 minus 0.40) / 0.80 f = (0.48 minus 0.40) / 0.80 f = 0.08 / 0.80 f = 0.10

Kelly tells you to bet 10% of your bankroll. If you have $1,000, you’d bet $100.

Now, 10% sounds aggressive for many bettors, and rightly so. Full Kelly assumes your probability estimates are perfect, and they never are. That’s why standard practice is to use Kelly fractions. Half Kelly (betting half what the formula says) and Quarter Kelly are the most popular among professionals. With Half Kelly, you’d bet $50 in the example above.

When to apply the Kelly Criterion?

Kelly works best when you have reasonable confidence in your probability estimates. If your model has shown to be calibrated over hundreds of bets, Kelly is the most efficient way to capitalize on that edge.

It’s especially useful when you bet across a wide range of odds. If all your bets are at similar prices, the difference between Kelly and a flat stake is small. But if one day you bet at 1.50 and another at 5.00, Kelly automatically adjusts the bet size based on the risk and reward of each situation.

Don’t use full Kelly if you’re starting out or your estimation history is short. The variance with full Kelly is brutal. An error in your probability estimate can lead to oversized bets that seriously damage your bankroll. Start with Quarter Kelly and scale up as you confirm your model’s accuracy.

It’s also not ideal for parlays. Kelly is designed for single bets. With parlays, the calculations get considerably more complex and selection interactions make the simple formula no longer optimal.

Practical example

You have a $2,000 bankroll and three bets for the Serie A weekend.

Bet 1: Juventus beats Torino. Odds 1.65, your estimated probability 68%. b = 0.65, p = 0.68, q = 0.32 f = (0.65 × 0.68 minus 0.32) / 0.65 = (0.442 minus 0.32) / 0.65 = 0.1877 = 18.8% Half Kelly: 9.4% = $188

Bet 2: Atalanta beats Lecce. Odds 1.45, your estimated probability 72%. b = 0.45, p = 0.72, q = 0.28 f = (0.45 × 0.72 minus 0.28) / 0.45 = (0.324 minus 0.28) / 0.45 = 0.0978 = 9.8% Half Kelly: 4.9% = $98

Bet 3: Lazio beats Empoli. Odds 1.90, your estimated probability 55%. b = 0.90, p = 0.55, q = 0.45 f = (0.90 × 0.55 minus 0.45) / 0.90 = (0.495 minus 0.45) / 0.90 = 0.05 = 5% Half Kelly: 2.5% = $50

Notice how Kelly makes you bet more on Juve (big edge) and less on Lazio (small edge). That’s exactly what you want: more money where you have more edge.

Common mistakes

1. Using full Kelly with imprecise estimates. It’s the fatal mistake. If your estimated probability is off by 5%, full Kelly will have you betting absurd amounts. Always start with Kelly fractions until you have real confidence in your model.

2. Not recalculating the bankroll. Kelly works on your current bankroll, not the initial one. If you started with $1,000 and now have $1,400, you should calculate percentages on the $1,400. The same if it drops to $800. Many bettors fix their bets on the initial bankroll and miss out on the compound growth advantage.

3. Applying Kelly to bets without edge. If the formula gives you a negative or zero result, don’t bet. Sounds obvious, but some modify their estimates to “let” Kelly tell them to bet. That’s manipulating numbers to justify an emotional decision.

4. Stacking Kelly fractions without adjusting. If you have five simultaneous bets each calling for 10% of bankroll, you can’t bet 50% total. You have to adjust proportionally to avoid overexposure. The general rule is your total exposure shouldn’t exceed 20-25% of your bankroll at any given moment.

Frequently Asked Questions

Does Kelly work for live betting?

Yes, the formula is valid for any bet where you can estimate a probability and have odds. The challenge in-play is that probabilities change rapidly and you need to calculate everything in real time. Some professional bettors use software that calculates Kelly automatically as the match unfolds.

What Kelly fraction do professionals recommend?

Most professional bettors I know use between Quarter and Half Kelly. The choice depends on how much confidence you have in your estimates and your variance tolerance. With Half Kelly you sacrifice 25% of optimal growth but cut variance in half, which for most is an excellent tradeoff.

Can Kelly be used for parlays and systems?

It can, but the simple formula is no longer valid. For parlays you need to calculate the joint probability of all results and the combined odds. There are Kelly extensions for correlated simultaneous bets, but they’re mathematically complex. For most bettors, it’s more practical to apply Kelly only to singles.