The Kelly Criterion: The Math Behind Optimal Bet Sizing

If you have ever wondered how much to wager on a bet you believe is favorable, or how much of your portfolio to allocate to a high-conviction investment, you have encountered one of the oldest and most important questions in the mathematics of risk. And it has a remarkably precise answer.

The Kelly Criterion, developed by Bell Labs physicist John Larry Kelly Jr. in 1956, is a formula that tells you exactly how much to bet to maximize the long-term growth rate of your bankroll. It is used by professional gamblers, hedge fund managers, venture capitalists, and anyone else who makes repeated decisions under uncertainty with real money on the line.

Understanding the Kelly Criterion will not make you a better handicapper or stock picker. It does not help you find edge. But once you have an edge, Kelly tells you precisely how to exploit it most efficiently, and equally importantly, how to avoid blowing up even when you are right about the odds.

The Formula

The Kelly Criterion for a simple bet with two outcomes (win or lose) is:

f* = (bp - q) / b

Where:

The formula looks simple, and it is. But its implications are profound, and misapplying it is one of the fastest ways to go broke.

A Worked Example

Suppose you have found a betting opportunity where you believe you have a 60% chance of winning an even-money bet. That means:

Plugging into the formula:

f* = (1 x 0.60 - 0.40) / 1 = 0.20

The Kelly Criterion says you should bet 20% of your bankroll on this wager.

That might seem like a lot. It might also seem too precise, like a false sense of mathematical certainty applied to an uncertain situation. Both intuitions are correct, and both point to why understanding Kelly requires going beyond the formula itself.

Why It Works: Maximizing Geometric Growth

The Kelly Criterion is not trying to maximize your expected profit on any single bet. If that were the goal, you should bet everything on any positive expected value opportunity. Obviously, that strategy leads to ruin the first time you lose.

Instead, Kelly maximizes the expected geometric growth rate of your bankroll. This is a critical distinction. Arithmetic growth is about how much you gain on average. Geometric growth is about how your bankroll compounds over time, accounting for the fact that a 50% loss requires a 100% gain to recover.

Consider two bettors, each with a 60% edge on an even-money bet. Bettor A bets 50% of their bankroll each time. Bettor B bets the Kelly-optimal 20%. After a large number of bets, Bettor B's bankroll will almost certainly be larger than Bettor A's, even though Bettor A is betting more aggressively. The reason is that Bettor A's large losses in the 40% of cases where they are wrong destroy capital faster than their large wins can rebuild it. Variance kills compounding.

The mathematical proof of this, rooted in information theory and the work of Claude Shannon, is elegant. But the intuition is straightforward: the Kelly Criterion finds the sweet spot where you are betting enough to take meaningful advantage of your edge, but not so much that the inevitable losing streaks devastate your bankroll.

The Danger Zone: What Happens Above Kelly

One of the most important insights from Kelly theory is that betting more than the Kelly amount is strictly worse than betting the Kelly amount, and it is worse in every meaningful sense.

Betting double Kelly, for instance, produces the same expected geometric growth rate as betting nothing at all. And betting anything above double Kelly actually produces a negative expected geometric growth rate, meaning you are almost certain to go broke over time, even when you have a genuine edge.

This is counterintuitive. Most people assume that if you have an edge, betting more should always be better. Kelly theory demonstrates decisively that this is false. Overbetting with a real edge is mathematically equivalent to or worse than having no edge at all. This single insight has saved countless professional bettors and investors from ruin.

Fractional Kelly: What Practitioners Actually Do

If full Kelly is mathematically optimal, why do most practitioners use half-Kelly or even quarter-Kelly?

The answer lies in the gap between theory and practice. The Kelly formula assumes you know your true edge with certainty. In reality, you never do. Your estimate of p, the probability of winning, is always an estimate, subject to error, bias, and incomplete information.

If your edge is smaller than you think, full Kelly will lead to overbetting. Since the consequences of overbetting are severe (as discussed above), it makes sense to build in a margin of safety. Fractional Kelly achieves this by betting a fixed fraction of the Kelly-recommended amount.

Half-Kelly (betting 50% of the Kelly amount) achieves about 75% of the growth rate of full Kelly while dramatically reducing the variance and drawdowns you experience. For most practitioners, this tradeoff is extremely attractive.

Quarter-Kelly reduces growth further but provides an even smoother ride. It is commonly used by investors and bettors who prioritize capital preservation or who have significant uncertainty about the accuracy of their edge estimates.

The general principle is: the less certain you are about your edge, the smaller the fraction of Kelly you should use. When your estimates are noisy, humility in bet sizing is mathematically rewarded.

Applications Beyond Betting

While the Kelly Criterion was developed in the context of gambling, its principles apply wherever you make repeated decisions about how much to risk.

Investment portfolio sizing is a natural application. How much of your portfolio should you allocate to a single stock you believe is undervalued? Kelly provides a quantitative framework for answering this question, replacing gut feeling with mathematical discipline. Many value investors, whether they cite Kelly explicitly or not, use portfolio sizing approaches that are conceptually equivalent to fractional Kelly.

Startup decision making involves similar tradeoffs. An entrepreneur deciding how much capital to invest in a new product line is making a bet. If they invest too little, they fail to capture the opportunity. If they invest too much and the product fails, it may threaten the entire company. Kelly thinking encourages sizing the investment proportionally to the estimated edge and the confidence in that estimate.

Venture capital is perhaps the most natural application of Kelly-style thinking outside of gambling. A venture fund makes many bets, each with a high probability of failure but a potentially massive payoff. The Kelly framework helps VCs think about portfolio construction, position sizing, and how to balance the number of investments against the size of each check.

Famous Users of the Kelly Criterion

Ed Thorp is the most famous Kelly practitioner. A mathematics professor who first used card counting to beat blackjack in the 1960s, Thorp explicitly used the Kelly Criterion to size his bets at the casino and later to manage his hedge fund, Princeton Newport Partners, which generated annualized returns of over 20% for nearly two decades with minimal drawdowns. Thorp's career is the most compelling real-world demonstration that Kelly-optimal sizing, combined with a genuine edge, produces extraordinary long-term results.

Warren Buffett has never publicly cited the Kelly Criterion, but his investment philosophy of making concentrated bets on high-conviction ideas is deeply consistent with Kelly thinking. Buffett has repeatedly argued against excessive diversification, stating that wide diversification is only required when investors do not understand what they are doing. This is, in essence, a Kelly argument: if your edge is large and well-understood, you should bet big.

Professional sports bettors and poker players represent another population of Kelly practitioners. The professional betting community has long understood that bankroll management is as important as handicapping, and the Kelly Criterion provides the theoretical foundation for the disciplined bet sizing that separates professionals from recreational bettors who may have similar analytical skill but inferior risk management.

Common Mistakes

Even people who understand the Kelly Criterion make predictable errors in applying it.

Overestimating your edge is the most common and most dangerous mistake. If you believe you have a 60% chance of winning when the true probability is 52%, you will bet far more than is optimal, potentially crossing into the danger zone above true Kelly. Always be conservative in your edge estimates.

Ignoring correlation is a subtler error. The Kelly formula as presented above assumes each bet is independent. If your bets are correlated, such as betting on multiple games in the same sport on the same day, or holding multiple stocks in the same sector, the effective risk is much higher than it appears, and your Kelly fraction should be lower.

Using full Kelly in practice is rarely advisable. As discussed, the gap between estimated and true edge means fractional Kelly is almost always the right approach. The mathematical optimality of full Kelly is a theoretical result that holds only when your inputs are perfectly accurate.

Applying Kelly to a single bet misunderstands its purpose. The Kelly Criterion optimizes long-run compounding. On any single bet, it has nothing special to say. Its power emerges only over many repeated decisions, where disciplined sizing allows your edge to compound without the interruption of catastrophic losses.

The Bottom Line

The Kelly Criterion answers one of the most important questions in any domain involving risk: given that I have an advantage, how aggressively should I exploit it? The answer is disciplined, proportional, and often more conservative than your instincts suggest.

Whether you apply it to sports betting, financial markets, or any other arena of repeated decision-making under uncertainty, understanding Kelly will make you a better risk manager and a more disciplined allocator of capital.

For a comprehensive treatment of the Kelly Criterion in the context of sports wagering, including worked examples across multiple sports and bet types, see the Sports Betting textbook. For an exploration of how Kelly-style thinking applies to prediction markets and probabilistic forecasting, see the Prediction Markets textbook. Both are free to read online.