The Kelly Criterion sizes your bet based on your edge and the offered odds. Bet too small and you under-compound; too large and you risk ruin even with positive expected value. This guide walks through Kelly, why full-Kelly is brutally volatile in practice, and how fractional Kelly is what professionals actually use.
The Kelly formula in plain English
Kelly fraction = (probability × decimal odds − 1) ÷ (decimal odds − 1). The output is the percentage of bankroll to bet.
Worked example: you estimate a 60% chance an event happens, and the book offers decimal odds of 2.0 (American +100). Kelly fraction = (0.60 × 2.0 − 1) / (2.0 − 1) = 0.20 / 1.0 = 20%. Kelly says bet 20% of bankroll.
20% feels enormous, and that's the problem with full Kelly in practice. A handful of losses in sequence on 20%-of-bankroll bets produces drawdowns most bettors cannot psychologically tolerate.
Fractional Kelly: what professionals actually use
Full Kelly is brutally volatile in practice. Most professionals use quarter-Kelly or half-Kelly — same long-run growth, far smaller drawdowns.
Quarter-Kelly on the worked example: 20% × 0.25 = 5% of bankroll. Same expected positive growth, drawdowns roughly 1/4 the magnitude of full Kelly. This is the standard for serious sports bettors.
Half-Kelly (10% on the worked example) sits in between — more growth, more drawdown. Suitable for bettors with high conviction in their edge estimate and stomach for variance.
Edge estimation: the input that breaks Kelly
Kelly only works if your edge estimate is accurate. Overestimating edge converts Kelly from a growth-maximizing tool into a bankroll-destruction tool. Underestimating edge under-compounds (less bad, but suboptimal).
If you're guessing edge from gut feel, flat-stake 1–2% of bankroll until you have a tracked sample of 500+ bets. Once you have data, your closing-line value (CLV) and win rate at known odds give you a real edge estimate.
Most recreational bettors over-estimate edge by 2–5x. Apply a discount factor to your gut estimate (multiply your perceived edge by 0.5) before running through Kelly. Better to under-compound than blow up.
When not to use Kelly
Kelly assumes infinite bankroll resilience and an accurate edge model. Recreational bettors usually lack both. For most retail bettors, flat-staking 1–2% of bankroll per bet is a safer default until you've genuinely measured your edge over a meaningful sample.
Arbitrage betting (covering all outcomes for guaranteed profit) doesn't use Kelly — the EV is fixed and risk-free, so stake size is governed by maximum book limits, not bankroll risk. See our value vs arbitrage comparison for the difference.
Applying fractional Kelly on Stake
Pre-calculate stake sizes before betting. Trying to compute Kelly in real-time during live betting is a recipe for emotional sizing.
Cap any single bet at 5% of bankroll regardless of what Kelly says. This is a hard ceiling that protects against edge mis-estimation on any individual bet.
Combine Kelly sizing with Stake's multi-bet bonus on 3+ leg parlays for additional EV layering. The multi-bet bonus boost is a real return on volume that compounds with disciplined sizing.
Questions readers ask about this guide
Is Kelly Criterion safe?
Full Kelly is theoretically growth-maximizing but produces 50%+ drawdowns historically. Quarter-Kelly is what most professionals use for risk-adjusted growth.
How do I know my edge?
Track every bet's offered odds and closing odds over 500+ bets. Closing-line value (beating the closing line) is the most reliable real-time edge signal.
Can Kelly be used for casino games?
Only on positive-EV opportunities (e.g., +EV promotions, advantage play). Negative-EV casino games have no Kelly solution — the optimal bet is zero.
Should I bet larger after winning?
Yes, in a Kelly framework. Your bankroll grows after wins, so the absolute size of each bet grows even though the percentage stays constant. This is the compounding mechanism.
Claim the full 200% up to $1,000 welcome bonus
One verified link. No promo code to remember. Bonus credited instantly on your first deposit.
