The Kelly Criterion

Once you've found a value bet, one question remains: how much do you stake? Bet too little and you leave growth on the table. Bet too much and variance ruins you. The Kelly criterion answers that question precisely — it's the staking method that maximises the long-term growth of your bankroll.

It's also the exact opposite of Martingale. Kelly stakes more when you have a bigger edge, never more because you just lost.

The formula

For decimal odds, the Kelly stake — as a fraction of your bankroll — is:

stake fraction = (probability × decimal odds − 1) ÷ (decimal odds − 1)

The top of that fraction should look familiar: probability × decimal odds − 1 is your edge, the EV per unit staked. If it's zero or negative, Kelly returns zero or less — its way of telling you there's no bet here. The bottom scales that edge by the odds. The bigger your edge, the bigger the stake; the longer the odds, the more cautious it gets.

A worked example

Say you've found a bet at 2.50 that your model rates a 50% chance:

• Edge = (0.50 × 2.50) − 1 = +0.25
• Kelly fraction = 0.25 ÷ (2.50 − 1) = 0.25 ÷ 1.50 = 0.167

Kelly says stake 16.7% of your bankroll. On a 1,000 bankroll, that's 167.

And that number should make you slightly uncomfortable — which is the right reaction.

Why full Kelly is too much

Sixteen percent on a single bet is a lot. Full Kelly is mathematically optimal only if your probability estimate is exactly right and you can stomach enormous swings. Two problems with that in real betting:

1. Your probabilities are estimates, not facts. If your 50% is really 46%, full Kelly is now overbetting badly — and overbetting is how Kelly turns dangerous.
2. The drawdowns are brutal. Even when your numbers are right, full Kelly regularly halves your bankroll on the way up. Most people can't hold a strategy through that.

The standard fix is fractional Kelly: multiply the recommended stake by a fraction — commonly a quarter or a half. Half-Kelly keeps about three-quarters of the growth with far less than half the volatility, and it forgives small errors in your estimates. For most bettors, quarter- to half-Kelly is the sensible home.

Calculate your stake

Enter the odds, your estimated win chance, your bankroll, and a Kelly fraction. Note how the recommended stake collapses to zero the moment your probability drops below the price's implied probability — no edge, no bet.

The one catch

Kelly is only as good as the probability you feed it. Garbage in, overbetting out. Its entire output hinges on that win-chance number being genuinely better than the bookmaker's — which is exactly the hard part of betting, and exactly what statistical models exist to solve.

Tofiko's models produce those probability estimates across hundreds of leagues. Feed a model probability and the live price into the calculator above, apply a sensible fraction, and you've turned a vague "this looks like value" into a specific, disciplined stake.