Kelly Criterion Betting: Stake Size, Formula, and Why Fractional Kelly Wins

The Kelly criterion answers one question: how much of your bankroll should you stake on a bet you believe has positive expected value? This page walks through the formula, a worked example, and why serious bettors almost never bet the full amount it recommends.

The Kelly criterion is a bet-sizing formula that returns the fraction of your bankroll to stake on a bet with positive expected value. Given your probability estimate p, decimal odds o, and q = 1 − p, the full-Kelly stake fraction is (bp − q) / b, where b = o − 1. It maximises long-run bankroll growth when your probability estimates are accurate. When they are not — and in sports betting they never are exactly — bettors use fractional Kelly, staking a quarter or half of the recommended amount to protect against estimation error.

What is the Kelly criterion?

The Kelly criterion is a mathematical formula for the optimal stake size on a bet with positive expected value. It tells you what percentage of your bankroll to risk on a single wager, given:

1. Your probability estimate
2. The odds you can get
3. Your current bankroll

It is named after John L. Kelly Jr., who derived the underlying mathematics in a 1956 Bell Labs paper, A New Interpretation of Information Rate. Kelly's original problem was about signal transmission over noisy channels; bettors use the same result to size stakes.

Kelly maximises expected logarithmic bankroll growth under its assumptions: accurate probabilities and independent bets. Those assumptions are demanding, so it is a sizing framework, not a result guarantee.

How do you calculate the Kelly criterion?

The full-Kelly stake fraction is:

Stake fraction = (bp − q) / b

Where:

• p = your probability estimate
• q = 1 − p (the opposite probability)
• b = decimal odds − 1 (the win return per unit staked at those odds)

The quantity bp − q is the expected return per unit staked — not simply the gap between your estimated probability and the implied probability.

Worked example. Take a match at decimal odds of 2.10. You estimate the outcome's true probability at 52%; the implied probability is 47.6%.

1. b = 2.10 − 1 = 1.10
2. q = 1 − 0.52 = 0.48
3. Kelly fraction = ((1.10 × 0.52) − 0.48) / 1.10 ≈ 8.4%

Full Kelly would suggest staking 8.4% of your bankroll on this bet. On an illustrative €10,000 bankroll, that is €840. In general, the stake is the Kelly fraction multiplied by your bankroll: if Kelly says 2% and your bankroll is €10,000, you stake €200.

What is full Kelly vs fractional Kelly?

Full Kelly means staking exactly what the formula returns. Fractional Kelly means staking a fraction of that — half, a quarter, or some other share you choose.

Using the worked example above:

Fraction	Share of bankroll	Stake on €10,000
Full Kelly (1/1)	8.4%	€840
1/2 Kelly	4.2%	€420
1/4 Kelly	2.1%	€210

The fraction is a risk choice, not a universal rule. Lower fractions reduce exposure to estimation error and correlated outcomes, at the cost of lower theoretical growth.

Why do professional bettors use fractional Kelly?

Because the formula is only as good as the probability you feed it — and your estimates are never exactly right.

The estimation problem in numbers. If you estimate a bet's probability at 52% but the true probability is 49%, your error is three percentage points — a realistic, small error. At 52% the expected return is +9.2%; at 49% it is +2.9%. The bet is not necessarily a loser — but it is a much smaller edge than the one you sized for. Full Kelly on the true 49% would stake 2.6% of bankroll, not the 8.4% you staked at 52%. A three-point miss more than triples your stake relative to what the true edge supports.

That is the case for fractional Kelly in one number. Professional bettors using EV+ should ask one question before staking: how much error tolerance do I need? Using 1/2 Kelly halves your stake, which reduces the damage of a three-point miss; using 1/4 Kelly reduces it further. The tighter your probability estimates, the closer to full Kelly you can afford to run. The wider your confidence bounds, the deeper into fractional Kelly you should go.

Kelly is not about win percentage. It is about long-run growth when you have an edge. A bet can be +5% EV and still lose 48% of the time — and losing small streaks are a normal part of that distribution. Kelly sizing ensures you are not wagering so much that a run of losses ruins you before the edge converges. A stake size you can hold through those streaks is worth more than theoretical growth you abandon after the first drawdown.

How much of your bankroll should you stake?

When you see an EV+ opportunity flagged at +5%, the edge metric already accounts for de-vigging and market consensus. Your job is deciding:

1. Do I believe this edge is real? (A 5% edge is smaller than a 10% edge; smaller edges have narrower confidence bounds.)
2. What staking fraction protects my bankroll?

A common rule of thumb:

• Strong, high-edge opportunities (8%+ edge, high confidence): 1/2 Kelly
• Moderate opportunities (3–7% edge): 1/4 Kelly
• Weak or low-confidence signals (1–3% edge): 1/8 Kelly or sit out

This is a heuristic, not advice and not a law. It depends on your bankroll size, how many simultaneous bets you carry, and your risk tolerance. The principle is consistent: use a fraction of full Kelly, and scale it with confidence in the edge. Pick a fraction and stick with it — do not flip to full Kelly when results are hot or dial down to 1/10 when you are cold. Staking strategy should be stable; only the bets themselves change.

Whatever fraction you choose, the final stake is the minimum of three numbers: the Kelly-calculated amount, the available liquidity at that book, and your internal position limit (step 6 in Building a staking strategy below).

Kelly criterion calculator

This article gives you the formula and the worked example, but the honest answer to "how much should I stake on this specific bet" is a calculation you run fresh each time — odds and your probability estimate change with every market. The fastest way to turn an EV+ opportunity into a stake size is the Kelly criterion calculator — enter the odds, your probability estimate, and your bankroll, and it returns the full, half, and quarter-Kelly stakes. Then apply the staking strategy below.

What are the limits of the Kelly criterion?

It needs an accurate probability input. Kelly is a sizing framework, not a result guarantee. It tells you how much to stake given a probability estimate; it cannot tell you whether that estimate is right. Garbage in, garbage out — the formula will happily size an oversized stake on a bad estimate.

It assumes independence. If you are staking five correlated bets on matches in the same league, Kelly calculated on each bet in isolation will oversize you relative to the true simultaneous risk. You either need to account for the correlation yourself or size more conservatively as a buffer.

It does not model execution risk. An apparent arbitrage still carries risk until every leg is accepted at the intended price: prices can change, bets can be rejected, and limits and liquidity can prevent the planned stake. Kelly is not the right sizing shortcut for that operational risk. Arbs are also constrained by limits and liquidity — you cannot always stake the full Kelly-calculated amount because one book's limit may not accommodate it. In those cases, stake as much as you can up to the Kelly number, then move to the next opportunity.

Kelly and bankroll management

Kelly sizing depends on your total bankroll, not just your available play capital. If you have €50,000 in total betting capital but only deploy €10,000 at a time, you should calculate Kelly based on €50,000, not €10,000.

A bet sized as 4% of a €10,000 active pool is actually 0.8% of a €50,000 total bankroll. That matters for long-term growth calculations and ruin probability. Most professional bettors track total bankroll — including inactive reserves — and size accordingly, which prevents accidental over-leveraging when multiple bets overlap.

Kelly staking is one part of a full bankroll management system, and the arithmetic above only compounds if you have enough sample — see variance and sample size for how long edges take to show up in results.

Building a staking strategy

A simple, robust staking approach:

1. Scan for opportunities on PhotonOdds (EV+, arbitrage, or dropping odds).
2. Filter by edge size and confidence level.
3. If you have an independently estimated probability, calculate Kelly as (bp − q) / b; do not substitute a display edge percentage without confirming its definition.
4. Apply a fractional Kelly multiplier (1/4 or 1/2, depending on edge size and your confidence).
5. Check the book's limit and available liquidity.
6. Stake the minimum of: (Kelly-calculated amount, available liquidity at that book, your internal position limit).
7. Log the bet, including the odds, stake, and your edge estimate.

Over time, compare logged prices, closing prices, and outcomes to test whether your assumptions need revisiting. Track those records in the Bet Log rather than treating a short run of results as proof.

18+ only. Betting carries risk. PhotonOdds provides analytical and educational tools, not a promise of profit or a recommendation to place a bet. If gambling is causing harm, see Responsible Gambling.