The Most Important Question

Finding an edge is only half the battle. The other half — and the one that destroys more careers than bad analysis — is knowing how much to wager. Bet too little and you leave profit on the table. Bet too much and a normal losing streak wipes you out. The Kelly criterion is the mathematical answer to this question.

Developed by John Kelly at Bell Labs in 1956, the Kelly criterion calculates the optimal fraction of your bankroll to bet given your edge and the odds offered. It maximizes the expected geometric growth rate of your bankroll. Over time, no other staking strategy produces higher terminal wealth.

The Formula

The simplest form of the Kelly formula for a single bet:

f = (bp - q) / b

Where:
f = fraction of bankroll to bet
b = decimal odds minus 1 (the net return per unit bet)
p = your estimated probability of winning
q = 1 - p, the probability of losing

Example: You estimate a team has a 55% chance to win, and the market offers -110 (decimal 1.909, so b = 0.909).

f = (0.909 * 0.55 - 0.45) / 0.909
f = (0.500 - 0.45) / 0.909
f = 0.05 / 0.909 = 0.055

Kelly says bet 5.5% of your bankroll. With a $10,000 bankroll, that's $550. You don't have to do this math each time — the Kelly Calculator takes your bankroll, your estimated probability, and the offered price, and returns the full-Kelly and fractional-Kelly stakes side by side.

Why Kelly Works

Kelly maximizes the expected logarithm of wealth. This sounds abstract, but the intuition is powerful: it scales bets proportionally to edge, penalizes overbetting extremely harshly, and never risks ruin in a single wager.

Consider what happens with a 60% edge on a coin flip at even money. Kelly says bet 20% of your bankroll. If you win, your bankroll grows 20%. If you lose, it shrinks 20%. The expected log wealth is maximized at this fraction. Betting 40% might seem attractive — bigger edge, bigger bet — but the increased risk of a drawdown more than cancels the expected gain.

This is the insight many bettors miss: the relationship between bet size and expected growth is not linear. Doubling your bet does not double your expected growth. Beyond the Kelly fraction, expected growth actually declines.

"Kelly is not just about maximizing returns. It is about not blowing up. The logarithmic wealth function has an infinite penalty for hitting zero."

Fractional Kelly in Practice

Full Kelly is too aggressive for most bettors. It assumes you know your edge precisely — you don't. It assumes you can handle 50%+ drawdowns psychologically — you probably can't. In practice, virtually all professionals use fractional Kelly.

Half Kelly (betting 50% of the full Kelly fraction) is the most common approach. It reduces volatility by roughly 50% while retaining approximately 75% of expected growth. A bettor using half Kelly still outperforms virtually every alternative staking strategy over time.

Quarter Kelly is even more conservative, appropriate for bettors still validating their edge or working in high-uncertainty markets. The reduction in expected growth is meaningful, but the peace of mind and reduced risk of ruin may be worth it.

Estimating Your Edge

The Kelly formula requires you to know your edge. This is the hard part. If you think you have a 5% edge but you actually have a 1% edge, full Kelly will eventually destroy your bankroll.

Approaches to estimating edge:

  • Model-based: Your model outputs a probability. Compare it to the market's implied probability (after removing vig). The difference is your estimated edge.
  • CLV-based: Track your CLV over a large sample. Your average CLV is a reasonable proxy for your expected edge on future bets.
  • Conservative estimates: When uncertain, estimate on the low side. Underestimating your edge and betting less is far better than overestimating it and betting too much.

Common Mistakes

Betting a fixed amount per game: Flat staking ignores edge. A 1% edge bet and a 10% edge bet get the same wager, which is obviously wrong.

Martingale and chase systems: Doubling after losses is mathematically guaranteed to fail. It accelerates bet sizes during losing streaks — the worst possible time.

Overbetting based on confidence: "I really love this one" is not a substitute for a calculated edge. Human confidence is poorly calibrated. Trust the math.

Simulating Kelly

Before committing real money, simulate your Kelly strategy. Run 10,000 iterations of your betting schedule with your estimated edge in the Variance Simulator. Look at the distribution of outcomes. How often do you see 50% drawdowns? How often do you go bust? If the tail risk is unacceptable, reduce your Kelly fraction until it is. Pair it with the Losing Streak Calculator to see how long a run of losses you should expect even when you're playing correctly — most bettors give up at the third deviation that the math says is routine.