Kelly Criterion: Optimal Bet Sizing and Its Limits

What it is

The Kelly criterion is a formula that determines the fraction of your bankroll to bet on a given wager to maximize long-term growth, assuming you have an accurate estimate of the true probability. Developed by John Kelly in 1956, it balances risk and reward: bet too little and you grow slowly; bet too much and you increase the risk of ruin. It is widely used in investing and sports betting, but its application requires precise inputs and discipline.

In sports betting, the Kelly criterion answers the question: "Given my edge, how much should I bet?" It does not guarantee profit—it only maximizes growth if your probability estimate is correct. If you overestimate your edge, the formula can recommend stakes that actually hurt your bankroll.

The math

The full Kelly formula for a single bet is:

f = (odds × p − 1) / (odds − 1)

where:

• f = fraction of current bankroll to bet
• odds = decimal odds offered
• p = your estimated true probability of winning

The numerator, odds × p − 1, is your expected value per dollar (edge). The denominator, odds − 1, is how many times you profit. The ratio tells you the optimal bet size.

Example with realistic numbers: decimal odds of 2.50 (American +150) and your estimated probability p = 0.45 (45%). The edge is (2.50 × 0.45) − 1 = 1.125 − 1 = 0.125, or 12.5%. Then f = 0.125 / (2.50 − 1) = 0.125 / 1.50 = 0.0833, or 8.33% of bankroll. With a $1,000 bankroll, the Kelly bet is $83.33.

Now consider the same odds of 2.50 but with a true probability of 0.35 (35%). Your edge would be negative: 2.50 × 0.35 − 1 = 0.875 − 1 = −0.125. The formula yields a negative f, which means no bet. But if you mistakenly estimate p = 0.45, you bet 8.33% despite having no edge or even a negative one. With repeated such bets, ruin becomes likely. Overestimation is the most dangerous pitfall: the formula is aggressive, and a small error in p can lead to a large recommended stake that destroys bankroll.

Worked example

Let's walk through a concrete scenario. You are considering a bet on a baseball game. The sportsbook offers American odds of +150, which converts to decimal odds of 2.50. You have done detailed analysis and believe the true probability of this outcome is 50%. The edge is: (2.50 × 0.50) − 1 = 1.25 − 1 = 0.25 (25%). Then f = 0.25 / (2.50 − 1) = 0.25 / 1.50 = 0.1667, or 16.67% of your bankroll. With a $1,000 bankroll, that is a $166.67 bet.

But suppose your analysis was wrong and the true probability is only 40%. You would bet 16.67% based on your flawed estimate, but the actual edge is negative: (2.50 × 0.40) − 1 = 1.00 − 1 = 0, so no edge. Yet you are risking 16.67% of your bankroll on a zero-expectation bet. If this happens repeatedly, your bankroll will decline due to the natural variance, and the overbetting accelerates losses. The Kelly criterion magnifies estimation errors.

To see the effect of overestimation, take odds of 2.00 (American +100). With true p = 0.50, edge = 0, so f = 0. But if you overestimate p = 0.60, edge = 0.20, f = 0.20 / (2.00 − 1) = 20% of bankroll. Betting 20% on a coin flip with no edge is a quick path to ruin: after a few losses, your bankroll shrinks dramatically.

When not to use it

• When you cannot estimate probabilities with high confidence. The Kelly criterion assumes your p is accurate. In sports, true probabilities are unknown and estimates are noisy. Even a small bias in your estimate can produce stakes that increase long-term risk instead of growth. If your edge is uncertain, use fractional Kelly (e.g., half or quarter) to reduce variance.
• When bet size is constrained by limits or account restrictions. Kelly often recommends large fractions (10–30%) that exceed maximum stakes or raise flags with sportsbooks. Account limits, reduced limits, or outright bans prevent you from placing the full Kelly amount, rendering the method impractical. You must adapt to the real constraints of the betting environment.
• When the market is inefficient due to liquidity or information. Kelly works best in efficient markets where odds reflect true probabilities minus margin. In illiquid markets (e.g., obscure leagues, prop bets), odds can be stale, distorted by a few bettors, or driven by sentiment. Estimating p becomes even harder, and Kelly amplifies errors.
• When you are at the beginning or end of your betting session. Kelly assumes you have many independent bets over time. If you only bet occasionally, the mathematical growth optimality does not apply over a small number of bets. Variance dominates, and a single large loss can be catastrophic. Only use Kelly when you have a large edge and a long sequence of bets.
• When you cannot handle the psychological volatility. Kelly stakes can swing wildly with each bet. A string of losses, even from positive-EV bets, can decimate a bankroll. Most bettors abandon the strategy after a few losses, leading to poor timing and emotional decisions. If you cannot stick to the plan through drawdowns, Kelly is not for you.

Use the Kelly criterion calculator to test different inputs and see how overestimation affects recommendations.

---

Educational content only. Sports betting involves financial risk and has a negative expected value in aggregate for the bettor. Must be 21+. If gambling ceases to be entertainment, seek professional help.