Strategy·Intermediate·7 min

Kelly Criterion: how much to stake

The mathematically optimal staking strategy — and why almost everyone bets half.

TL;DR

The Kelly Criterion is a formula for staking that maximises long-term bankroll growth. The full Kelly fraction is:

f = (p × (b + 1) − 1) / b

where:

p = your estimated true probability
b = decimal odds − 1 (net odds)

Bet f as a fraction of your bankroll. Negative or zero f means don't bet.

Worked example

You think Liverpool is 57% to win at 1.90 odds.

p = 0.57
b = 1.90 - 1 = 0.90

f = (0.57 × 1.90 - 1) / 0.90
f = (1.083 - 1) / 0.90
f = 0.092

→ Stake 9.2% of your bankroll. On a €1000 bankroll, that's €92.

Why Kelly is brutal in practice

Full Kelly is mathematically optimal only if your probability estimate is exactly right. If you over-estimate your edge by 20% (very common), Kelly stakes can wipe you out fast. The variance is also nasty — drawdowns of 30–50% during normal losing runs.

What the pros actually do

Fractional Kelly — bet a fraction (commonly 0.25× or 0.5×) of what Kelly says. This trades some long-term growth for huge variance reduction.

Quarter Kelly on the Liverpool example:

0.092 × 0.25 = 2.3% of bankroll

Much more bearable. Still aligned with the math. The +EV scanner on strikey.io reports half-Kelly (0.5×) by default in the "Kelly $100" column — read as "% of your bankroll" if you scale up.

Common Kelly mistakes

Using book odds for p. p should be your estimate of the true probability, not what's implied by the price. If p = 1/odds, Kelly always returns zero. Devig sharps' prices to get a defensible p (see devigging guide).
Kelly on parlays. Kelly assumes a single bet. Multi-leg correlated parlays break the math. Stake parlays at a small fixed fraction (1–2% of bankroll) instead.
Re-Kelly-ing during a losing streak. Kelly recomputes off your *current* bankroll. If you fall from €1000 to €700, your next Kelly stake should be smaller. People who keep staking off the original bankroll go bust.

When to skip Kelly entirely

Small edges (<2%). The fraction will be tiny anyway; flat-staking 1u keeps things simple.
Heavily uncertain probabilities. If you're guessing p with ±5 percentage points of uncertainty, your Kelly stake is meaningless.
Books that limit you. Kelly assumes you can stake whatever the formula says. If they cap you at €20, just bet €20.