How to Think About What Kelly Criterion Actually Tells You About Bet Sizing

Most gambling coverage focuses on whether a bet has positive or negative expected value. That question matters. But there is a second question worth asking even after you have answered the first: if a bet does favor you, how much of your bankroll should you actually put on it? Bet too little and you leave value on the table over time. Bet too much and a short run of bad luck can wipe you out before the math has any chance to work in your favor. The Kelly Criterion is a formula designed to answer exactly that question.

What the Formula Is

The Kelly Criterion, developed by physicist John L. Kelly Jr. in 1956, produces a fraction of your bankroll to wager on a single bet. The formula in its simplest form is:

Kelly fraction = (bp - q) / b

Where: - b is the net odds you receive on the bet (what you win per unit wagered) - p is the probability that the bet wins - q is the probability that the bet loses (which equals 1 - p)

That is it. The formula is short. The reasoning behind it, however, is worth unpacking carefully.

A Worked Hypothetical

Imagine a hypothetical coin flip where the coin is slightly biased. You have good reason to believe it lands heads 55 percent of the time. A bookmaker is offering even money: you win one unit for every unit you stake, and you lose your stake if it comes up tails.

Plugging into the formula: - b = 1 (even money pays one unit per unit wagered) - p = 0.55 - q = 0.45

Kelly fraction = (1 x 0.55 - 0.45) / 1 = (0.55 - 0.45) / 1 = 0.10

The formula says to bet 10 percent of your current bankroll on this flip. Not a fixed dollar amount, but a fraction of whatever you currently have.

If you started with 100 units and won the first bet, your bankroll grows to 110 units. On the next flip you would bet 10 percent of 110, which is 11 units. If you then lost, you would have 99 units. The bet size adjusts continuously as the bankroll changes.

Why the Fraction, Not a Fixed Amount

The reason Kelly sizing uses a fraction rather than a fixed stake is that it is trying to maximize the long-run growth rate of your bankroll, not the expected value of any single bet. Those two goals sound similar but they produce different answers.

Expected value cares only about the average outcome across many bets. If your edge is real, betting your entire bankroll every time maximizes expected value in the short run but guarantees eventual ruin because a single loss ends the game. Kelly sizing weights growth rate against the risk of ruin explicitly. The fraction it produces is the one that, under its assumptions, grows your bankroll fastest over a long sequence of identical bets without driving the bankroll to zero.

What Happens When You Bet More Than Kelly Suggests

Over-betting relative to the Kelly fraction is a concept worth understanding. If the Kelly fraction for a bet is 10 percent and you bet 20 percent, you are betting twice the Kelly amount. This is called "overbetting" in this framework. Mathematically, overbetting above the Kelly fraction reduces long-run growth rate, and above twice the Kelly fraction it produces negative expected growth even if each individual bet has positive expected value. The edge exists but the bet sizing destroys it over many rounds.

This is one reason the Kelly Criterion is sometimes called the "growth-optimal" criterion. Any fraction larger than the Kelly fraction reduces long-run growth compared to what Kelly would produce.

The Assumptions You Must Check

The Kelly Criterion only works correctly when its assumptions hold, and it is important to know what those assumptions are.

First, you must know p accurately. In most casino games, p is defined by the rules and is known exactly. In sports betting, p is your estimate, and if that estimate is wrong the Kelly fraction is wrong. An overconfident estimate of your edge produces an inflated Kelly fraction and higher real risk than you intend.

Second, the formula in its basic form assumes a single bet type with two outcomes: win or lose. It can be extended to bets with multiple outcomes, but the arithmetic becomes more involved.

Third, Kelly sizing assumes you will keep betting the same game repeatedly and that your bankroll is the only relevant resource. In practice, you may have other financial obligations, a fixed session budget, or limited time. Those real-world constraints can make a smaller fraction than Kelly suggests more appropriate.

Half-Kelly and Fractional Kelly

Because the Kelly fraction is sensitive to errors in your estimate of p, many practitioners of quantitative betting use what is called fractional Kelly, most commonly half-Kelly. This simply means betting half of whatever the full formula produces. In the hypothetical above, that would be 5 percent instead of 10 percent.

Half-Kelly reduces the long-run growth rate compared to full Kelly, but it also reduces the volatility of your bankroll path considerably. For anyone uncertain about the accuracy of their edge estimate, which in practice covers nearly all situations, fractional Kelly provides a margin of safety without abandoning the framework entirely.

Where This Does and Does Not Apply

The Kelly Criterion applies most cleanly in situations where you genuinely have a positive expected value edge. For the vast majority of standard casino games, the house edge means p is against you by construction, and the Kelly formula correctly produces a negative number or zero, which means the mathematically correct bet under this framework is zero.

The formula is more relevant in contexts where an edge might theoretically exist, such as certain card games under specific conditions, sports betting where your probability estimate differs meaningfully from the line's implied probability, or poker where skill can create a measurable edge over time. Even in those contexts, accurately estimating your true edge is difficult, which is why the assumptions section above is not optional reading.

The Practical Takeaway

The Kelly Criterion gives you a principled way to connect the size of your estimated edge to the size of your bet. It does not create an edge where none exists, and it does not guarantee you will win. What it does is show that bet sizing is not arbitrary. The fraction you choose has mathematical consequences for long-run growth and short-run volatility, and those consequences can be calculated rather than guessed. Understanding the formula helps you see that betting more is not always better, and that the relationship between your edge, your bankroll, and your bet size is one worth thinking through before you play.