Kelly Criterion Sports Betting — Optimal Stake Sizing Guide 2026

Most bettors who understand value betting still blow up their bankrolls — not because their picks are wrong, but because their stakes are wrong. They win 58% of bets at odds of 2.00 and still lose money, or they go bust on a 12-game losing streak that probability theory predicted was coming. The Kelly Criterion is the mathematical answer to the question "how much should I bet?" — and understanding it will change how you approach every single wager.

What is the Kelly Criterion and Why Does It Work Mathematically?

The Kelly Criterion was developed by mathematician John Larry Kelly Jr. at Bell Labs in 1956. His original paper, "A New Interpretation of Information Rate," was about signal noise in telephone networks — he never mentioned gambling. Ed Thorp, the blackjack legend and hedge fund pioneer, adapted it for betting and investing in the 1960s.

Kelly's key insight: if you have a genuine edge over a series of bets, there exists an optimal fraction of your bankroll to stake that maximises the long-run growth rate of your capital. Bet too little and you grow slowly. Bet too much and variance destroys you even when you're winning on paper. Kelly found the exact mathematically optimal point between those extremes.

The criterion maximises E[ln(wealth)], the expected logarithm of wealth. Logarithmic utility reflects a fundamental truth: losing 50% of your bankroll requires a 100% gain just to break even. The mathematics naturally incorporates this asymmetry. A bettor following Kelly will, with probability 1, outperform any other fixed-fraction strategy over a sufficiently long series of independent bets — that's not a guideline, it's a proven theorem.

The Kelly Criterion Formula — Every Variable Explained

f* — the Kelly fraction, expressed as a decimal proportion of your current bankroll. f* = 0.045 means stake 4.5% of your current bankroll.

b — the net decimal odds. For bookmaker odds of 2.10, b = 2.10 − 1 = 1.10. This is how much profit you collect per unit staked if you win.

p — your estimated true probability of winning. This is NOT the bookmaker's implied probability (which includes their margin). This is your model's estimate of the real-world likelihood of the outcome occurring.

q — the probability of losing, always equal to 1 − p. If you estimate p = 0.52, then q = 0.48. Note that in a two-outcome market p + q = 1, but in three-outcome football markets (home/draw/away), you'd need to account for each leg separately.

A simplified but equivalent form: f* = (edge / odds), where edge is your expected value expressed as a fraction. This makes intuitive sense: higher edge → larger stake; higher odds (more volatile outcome) → smaller fraction for the same edge.

Why Full Kelly Is Dangerous in Practice — The Variance Problem

Here's where theory meets brutal reality. Full Kelly assumes you know your edge precisely. In sports betting, you never do. Your EV estimate might be +5% but it carries enormous uncertainty — maybe the true edge is +2% or +8%, or even negative on a given bet. The Kelly formula is extremely sensitive to this: if your true edge is half what you estimated, Full Kelly stakes are still twice the mathematically optimal level.

Consider a concrete example. You have a 52% model edge at even money (odds 2.00), giving a Kelly stake of 4% of bankroll per bet. Over 500 bets, Full Kelly maximises growth — in theory. But the standard deviation of outcomes over any 100-bet stretch means you will regularly experience 15-bet losing streaks. With 4% Kelly stakes, 15 consecutive losses removes 46% of your bankroll: $10,000 drops to $5,421. That's before any model error is factored in.

The academic literature on Kelly demonstrates that "over-betting" — staking more than Kelly — always leads to eventual ruin given enough time. Betting 2× Kelly has negative expected logarithmic growth. Betting between 1× and 2× Kelly reduces growth rate but still avoids ruin. This creates an asymmetry: the cost of under-betting is smaller growth; the cost of over-betting can be catastrophic loss.

Quarter-Kelly (25%) — Why EVBets Uses This as the Standard

Quarter-Kelly means staking exactly 25% of the full Kelly fraction. If Kelly says 4% of bankroll, Quarter-Kelly says 1%. This isn't conservative — it's calibrated. Research by Thorp and others shows Quarter-Kelly delivers approximately 56% of Full Kelly's long-run growth rate while experiencing roughly 6% of the variance. That asymmetry is remarkable.

EVBets adopts Quarter-Kelly for every signal for three specific reasons. First, our EV estimates carry inherent uncertainty — the closing line validation gives us confidence in direction, but not precision in magnitude. Second, sports bettors face a real-world constraint full Kelly ignores: book limits. If you're betting 4% of a $50,000 bankroll at Pinnacle, you'll hit limit on many lines; 1% solves this. Third, psychological sustainability matters. A bettor who never experiences more than a 15% drawdown will stick to the strategy; one who sees 40% swings will deviate at the worst possible moment.

Half-Kelly (50%) is a reasonable alternative for bettors with high confidence in their edge estimates and longer time horizons. Full Kelly should only ever be used when edge is extremely well-established — think professional card counters with thousands of documented hands, not sports bettors with a 200-bet track record.

Practical Example: Real Numbers on a Real Bet

Let's walk through a complete calculation. EVBets flags a bet on a Champions League match: the home team to win at decimal odds of 2.10 at a soft bookmaker. Our no-vig Pinnacle model estimates the true probability of the home win at 52.4%.

The EV figure — +$1.14 on a $22.80 stake — represents a +5% edge. Over hundreds of bets at this edge, the compounding effect of Quarter-Kelly becomes significant without the catastrophic risk of full sizing.

Flat Betting vs Kelly vs Quarter-Kelly — 100 Bets, Same Edge

Using a simulation with 5% EV edge at odds 2.00 (52.5% win rate) over 100 bets, starting bankroll $1,000. Results shown as median outcomes and worst-5th-percentile outcomes across 10,000 simulations:

The table tells the story clearly. Full Kelly produces the highest median outcome but the worst 5th percentile — meaning 5% of bettors using it will have their $1,000 drop below $510 despite having a genuine edge. Quarter-Kelly's 5th percentile stays at $930. That's the difference between a bad month and a destroyed bankroll.

When Kelly Does NOT Apply

The Kelly Criterion requires three conditions that aren't always met in sports betting:

Frequently Asked Questions