How Much to Bet: The Kelly Criterion

Say you've found a genuinely good bet. A coin that lands heads 60% of the time, and you get to bet on heads, winning even money. You will come out ahead if you play long enough. But here's the question almost nobody asks and everybody gets wrong: how much of your money should you put on each flip?

Bet too little and you barely grow your pile despite having a real edge. Bet too much and a bad streak wipes you out before your edge can save you. There's a sweet spot in between, and in 1956, a physicist at Bell Labs named John Kelly found the exact formula for it, using ideas he borrowed from the science of telephone lines.

Why "bet everything on a sure edge" is a trap

Your instinct with a 60% coin might be to bet big. After all, the odds are in your favor. But watch what happens if you bet everything every time: the moment a single tails shows up, and it will, 40% of the time, you lose it all. Game over. One loss and you can never recover, because you have nothing left to bet.

So betting 100% is suicide even with a great edge. That reveals something important: the goal isn't to maximize your gain on any one bet, it's to survive and keep compounding. A wiped-out gambler earns nothing forever after, no matter how good the next opportunity is.

This is the deep insight. Money grows by multiplying, not adding. If you lose 50% and then gain 50%, you are not back to even, you're down to 75% of where you started. Losses hurt your compounding more than equal-sized gains help it. That asymmetry is why bet size matters so enormously.

Kelly's answer: bet in proportion to your edge

Kelly's formula tells you the fraction of your money to wager. You don't need the algebra to grasp the rule it produces:

• The bigger your edge, the more you bet. A 60/40 coin justifies a bigger bet than a 52/48 coin.
• The worse the payout odds, the less you bet. If a win pays little but a loss costs a lot, you shrink your bet.
• You never bet your whole stack on anything short of a literal sure thing.

For the 60% even-money coin, Kelly says to bet 20% of your money each time. Not 100%, not 5%, 20%. Bet that fraction every round and, over many rounds, your money grows faster than under any other betting scheme. That's the theorem: Kelly betting maximizes the long-run growth rate of your wealth.

The surprising origin: telephone wires

Here's the fun part. Kelly wasn't a gambler. He worked at Bell Labs on how much information you can push through a noisy telephone line, a field his colleague Claude Shannon had just invented. Kelly noticed that the math describing "how fast can information flow down a wire" is exactly the same math as "how fast can a gambler's money grow." A tip about a horse race is really just information arriving over a noisy channel, and the value of that information sets how fast you can grow your bankroll.

That's why the paper has such an unlikely title, "A New Interpretation of Information Rate." It's an information-theory paper that turned out to be the definitive guide to bet sizing. The connection is genuinely beautiful: your betting edge is a form of information, and information has a precise cash value.

How it's used today

Kelly stayed a cult idea among gamblers for decades, famously used by blackjack card counters and sports bettors, before serious investors embraced it. Legendary names like Ed Thorp (who beat both the casino and Wall Street) and Warren Buffett's partner Charlie Munger have pointed to Kelly-style thinking as central to how they size their bets.

In quant finance, the same logic drives position sizing: given a strategy's edge and riskiness, how much capital should you allocate to it? A strong, reliable signal earns a bigger slice; a weak or noisy one earns a sliver. The core Kelly wisdom, let the size of your bet scale with the strength of your edge, and never bet so big that a losing streak ruins you, is baked into how professional trading desks manage risk.

The honest limitations

Kelly's math is exact, but exactness depends on inputs the real world won't give you cleanly:

• You never truly know your edge. The formula assumes you know the real probabilities. Guess your edge too high and Kelly tells you to bet way too much, and overbetting is catastrophic in a way underbetting never is. Because of this, most practitioners deliberately bet a fraction of what Kelly says (like "half-Kelly"), trading a bit of growth for a much smoother, safer ride.
• Full Kelly is a wild ride. Even when it's mathematically optimal, the bankroll swings are stomach-churning, routine drops of 50% along the way. Most humans can't stomach that emotionally, and most institutions can't tolerate it professionally.
• It optimizes for the very long run. Kelly maximizes growth over many bets. If you only get a handful of shots, or you have a hard deadline, its logic bends and a more cautious size can be wiser.
• It assumes you can always re-bet your winnings. In markets with limited liquidity or one-off opportunities, the neat "compound forever" picture breaks down.

The practical takeaway from all this: Kelly is best used as a ceiling and a compass, not a literal instruction. It points you toward the right direction, bigger edge, bigger bet, and warns you loudly when you're about to bet dangerously large.

The one-line takeaway

Kelly proved that the smartest bet size grows with your edge but never risks ruin, because wealth compounds by multiplying, surviving to bet another day matters more than maximizing any single wager, and even a great edge can bankrupt you if you bet it too big.