Kelly Criterion Explained: How to Size Sports Bets for Maximum Growth

By BettingCalcs Editorial Team  |  Updated June 2026

Most sports bettors size their bets by gut feeling, arbitrary flat units, or whatever they can afford to lose on a given Sunday. The Kelly Criterion offers something fundamentally different: a mathematically derived answer to the question of how much to bet on each wager that maximizes long-run bankroll growth. It was developed not by a gambler, but by a physicist — and its principles are used today by professional sports bettors, quantitative hedge funds, and algorithmic traders. This guide explains the formula in full, shows you exactly how to apply it, and explains why most professionals deliberately use a fraction of it.

1. What the Kelly Criterion Is

The Kelly Criterion is a mathematical formula developed by physicist John Kelly Jr. at Bell Labs in 1956. Kelly was originally studying signal transmission problems, but his insight about optimal bet sizing under uncertainty has proven remarkably durable across applications from blackjack to venture capital to sports gambling.

The core idea is deceptively simple: for any bet where you have a genuine edge — where your estimated probability of winning exceeds the probability implied by the odds — there is a specific optimal fraction of your bankroll to risk. Bet more than that fraction, and you are taking on more variance than the edge justifies. Bet less, and you are leaving compounding growth on the table. Bet the Kelly fraction precisely, and you maximize the geometric growth rate of your bankroll over many bets.

The Kelly Criterion doesn't tell you whether you have an edge — that's your job. What it tells you is: given that you believe you have a specific edge, here is the mathematically optimal amount to risk. Applied correctly over thousands of independent bets, no fixed-fraction staking system produces faster bankroll growth. Applied incorrectly — particularly with inflated edge estimates — it produces catastrophic losses faster than nearly any other staking system.

2. The Kelly Formula

The full Kelly formula is:

Let's walk through a complete example using standard -110 American odds, the default for NFL point spreads:

• American odds of -110 convert to decimal odds of: (100 ÷ 110) + 1 = 1.909
• Net decimal odds: b = 1.909 − 1 = 0.909
• You estimate you have 55% probability of winning: p = 0.55, q = 0.45
• f* = (0.909 × 0.55 − 0.45) / 0.909
• f* = (0.500 − 0.45) / 0.909
• f* = 0.050 / 0.909
• f* = 0.055 = 5.5% of bankroll

So if you have a $1,000 bankroll and you genuinely have a 55% edge on a -110 line, the Kelly Criterion says to bet $55. That is the mathematically optimal bet size for maximizing your bankroll's long-run growth rate at these odds and edge. If you have an edge at all — if p > q/b — Kelly gives a positive f*, and that's your bet. If you have no edge (or negative edge), Kelly returns zero or a negative number: don't bet.

3. Why Full Kelly Is Too Aggressive in Practice

Kelly's formula is optimal under one critical condition: you know your exact edge. In academic applications, this assumption is sometimes reasonable. In sports betting, it is never fully met. This is where full Kelly becomes dangerous.

Consider what happens when your edge estimate is off. In the example above, you estimated 55% win probability. But what if your model is imprecise and your actual win rate is 52%?

• At 52% win probability and -110 odds: f* = (0.909 × 0.52 − 0.48) / 0.909 = (0.473 − 0.48) / 0.909 = −0.007/0.909 = negative Kelly (don't bet)
• You were betting 5.5% of your bankroll on a bet that had a negative expected edge.

The problem is systematic: bettors almost universally overestimate their edge. Confidence in one's model tends to exceed the model's actual accuracy. The Kelly Criterion is highly sensitive to this overestimation — if you believe your edge is 5.5% Kelly when it's actually 0%, you are aggressively over-betting zero-edge wagers and your bankroll decays instead of growing.

4. Fractional Kelly: The Industry Standard

The solution used by virtually all serious sports bettors is fractional Kelly: apply a fraction of the full Kelly output rather than the full amount. The most common variants:

The variance-reduction mathematics are compelling. Half-Kelly:

• Reduces expected log-wealth growth by approximately 25%
• Cuts the variance of outcomes by approximately 75%
• Reduces maximum expected drawdown from ~50% to roughly ~25%

Sacrificing 25% of expected growth to cut variance by 75% is an excellent tradeoff for anyone who doesn't have perfect edge estimation — which is everyone. This is why half-Kelly to quarter-Kelly is the standard among professional sports bettors and quantitative betting shops. The reduced growth rate is a small price to pay for staying solvent through the inevitable rough stretches that variance produces.

5. Worked Example With a Realistic Betting Scenario

Let's run a complete Kelly calculation for an NFL game with a plus-money line — the type of scenario where Kelly sizing can recommend surprisingly large bets:

Scenario: Bankroll $1,000. An NFL underdog is available at +150 American odds. Your model gives the team a 45% win probability (implied by +150 odds is approximately 40%).

• American +150 converts to decimal odds: (150 ÷ 100) + 1 = 2.50
• b = 2.50 − 1 = 1.50
• p = 0.45, q = 0.55
• f* = (1.50 × 0.45 − 0.55) / 1.50 = (0.675 − 0.55) / 1.50 = 0.125 / 1.50 = 0.083 = 8.3% full Kelly

Kelly says bet $83 on a $1,000 bankroll. Is that a lot? It depends on whether the 45% estimate is accurate.

Applying the fractional variants:

• Half-Kelly: 8.3% × 0.5 = 4.15% → $41.50 bet
• Quarter-Kelly: 8.3% × 0.25 = 2.075% → $20.75 bet

Now consider the edge sensitivity. If your model is slightly wrong and the team's true probability is only 41% (rather than 45%), your edge drops dramatically. At 41% true probability: f* = (1.50 × 0.41 − 0.59) / 1.50 = (0.615 − 0.59) / 1.50 = 0.025/1.50 = 1.7% — a legitimate bet, but much smaller. At quarter-Kelly, you'd bet $4.25. At full Kelly with the inflated 45% estimate, you'd have bet $83 on what was actually a near-zero-edge bet. Fractional Kelly protects you from that exact scenario.

6. Kelly Assumptions and Where It Breaks Down

The Kelly Criterion makes several mathematical assumptions that don't always hold in real-world sports betting:

Independence of bets: Kelly assumes each bet is independent of the others — the outcome of one bet doesn't affect the next. This is reasonable for individual game bets but breaks down completely with parlays, where the legs are often correlated (e.g., two games played in the same weather conditions, or a team's spread and total). Kelly does not extend cleanly to parlay betting.

Accurate probability estimates: As discussed, the formula is highly sensitive to estimation error. If you systematically overestimate your edge across all bets, Kelly will systematically over-size every bet. A well-calibrated model (where your 60% confidence picks actually win 60% of the time) is a prerequisite for Kelly to work as intended.

Reinvestment of winnings: Kelly assumes you continuously adjust your bet size as your bankroll grows or shrinks. If you win $50 and your bankroll goes from $1,000 to $1,050, your next Kelly bet should be based on $1,050, not $1,000. This requires consistent bankroll tracking and recalculation.

Infinite time horizon: Kelly's optimality is proven in the limit of many bets. With a small sample of bets, variance dominates. The expected value advantage of Kelly over flat betting takes hundreds or thousands of bets to materially manifest — a time horizon most recreational bettors don't operate on.

7. Practical Bankroll Management Rules

Whether or not you use Kelly, these hard limits protect your bankroll from the mistakes that end betting careers:

Hard cap at 5% per bet: Regardless of what the Kelly formula outputs, never risk more than 5% of your total bankroll on any single event. A Kelly recommendation above 5% means either (a) you have an extraordinarily rare edge — which should make you double-check your model — or (b) your probability estimate is inflated. Either way, 5% is the ceiling.

Treat Kelly > 10% as a red flag: Real edges large enough to justify 10%+ Kelly bets are essentially nonexistent in liquid betting markets. If your formula returns 10%+, the most likely explanation is model error or a data input mistake, not a genuine 10% edge opportunity. Use it as a prompt to recheck your math.

Track everything in a spreadsheet: Record every bet with: date, sport/game, your odds at bet time, your estimated win probability, the Kelly fraction you used, and the outcome. After 200+ bets, calculate your actual win rate and compare it to your estimated win rates. This comparison is the only honest way to know whether your edge estimates are calibrated to reality or wishful thinking.

Recalibrate if you're wrong: If your actual win rate is consistently below your estimated win rates over a meaningful sample, reduce your unit size and revisit your model's inputs. The bettors who survive long-term are those who respond to data rather than defending their original confidence. The Kelly Criterion is only as good as the probabilities you feed it.