The Kelly Criterion: How Much Should You Bet?
Finding a good bet is only half the job. The other half — the half that quietly decides whether you compound or go broke — is how much to stake. Bet too little and you leave growth on the table; bet too much and one cold streak wipes you out even when you're right on average. The Kelly criterion is the math that answers this, and it's the sizing framework most serious bettors and traders anchor to. Here's how it works and how to use it without blowing up.
What the Kelly criterion is
Kelly is a formula that outputs the fraction of your bankroll to wager on a bet, given two inputs: your estimated probability of winning and the odds you're being paid. It's the stake size that maximizes the long-run growth rate of your bankroll. Bet more than Kelly and your growth rate actually falls while your risk of ruin climbs; bet less and you grow more slowly but sleep better. It is not a betting system that picks winners — it assumes you already have an edge and tells you what to do with it.
The formula
For a simple win/lose bet, the Kelly fraction is:
Kelly fraction (f)
f = (b × p − q) / b
where b = the payout odds in decimal-minus-one terms (net profit per $1 staked if you win), p = your estimated probability of winning, and q = 1 − p (probability of losing).
A cleaner way to say it: Kelly ≈ your edge divided by the odds. The bigger your edge and the longer the price, the more you bet.
A worked example
Suppose you think a coin-flip-looking market is really 55% to hit, and it's priced at even money (a bet where you win $1 for every $1 risked, so b = 1). Plug it in:
- p = 0.55, q = 0.45, b = 1
- f = (1 × 0.55 − 0.45) / 1 = 0.10
Full Kelly says stake 10% of your bankroll on this bet. On a $2,000 bankroll that's a $200 wager. Notice how a modest-sounding 5-point edge (55% vs the 50% the even-money price implies) translates into a chunky 10% stake. That's the first warning sign: Kelly gets aggressive fast.
Now change the price. Same 55% read, but you get it at plus-money — say $1.50 profit per $1 (b = 1.5). Then f = (1.5 × 0.55 − 0.45) / 1.5 = 0.375 / 1.5 = 0.25. A better price on the same edge quadruples the recommended stake to 25% of bankroll. That is a lot of money on one bet — which is exactly why almost nobody bets full Kelly.
Why full Kelly is too aggressive
Full Kelly is mathematically optimal only if your probability estimate is exactly right. In real betting it never is. You're estimating 55% but the truth might be 52%, or 48%. Full Kelly is also brutally volatile: even with a real edge, a full-Kelly bankroll routinely suffers 50%+ drawdowns that most people can't stomach and that can end a bankroll if they hit early. The math rewards the long run, but you only get the long run if you survive the short one.
Sharps use fractional Kelly
The standard fix is to bet a fraction of the Kelly number — typically half Kelly or quarter Kelly. Half Kelly captures roughly 75% of the growth rate for about half the volatility, which is a fantastic trade. In the example above, full Kelly said 10%; half Kelly is 5%, quarter Kelly is 2.5%. Most disciplined bettors live in that 2–5% zone and treat the full-Kelly figure as a ceiling they never actually bet.
Kelly only works if your edge is real
Here's the part fake cappers never mention: Kelly is only as good as your probability estimate. The formula takes your win probability as gospel. If you feed it a fantasy — "I'm 60% here" when you're actually a 50% coin flip — Kelly will confidently tell you to bet a large slice of your bankroll straight into a negative-expectation wager, and the sizing turns a leak into a hemorrhage.
That's why Kelly can't be separated from expected value. You need a genuine, evidence-based reason to think your number beats the market's number before Kelly means anything. And in practice the cleanest proof that your edge is real is closing line value — consistently getting a better price than the market closes at. Positive CLV over a large sample is the signal that your win-probability inputs are honest; without it, any Kelly stake is just confident guessing.
The danger of over-betting a wrong edge
Betting above full Kelly is strictly worse in every way — lower growth AND higher risk of ruin. And since almost everyone overestimates their own edge, betting full Kelly on your stated number is usually already over-betting the true number. The safe posture is to be pessimistic about your edge and conservative with the fraction. When in doubt, bet less. You cannot go broke at quarter Kelly on a genuine edge; you absolutely can at full Kelly on an imaginary one.
Using it in practice
- Estimate your true win probability honestly — and then shade it toward the market, because the market is usually smarter than you think.
- Compute the full-Kelly fraction from that probability and the price on offer.
- Bet a fraction of it — half or quarter — never the full number.
- Recompute off your current bankroll as it grows or shrinks, so stakes scale with what you actually have.
- Track CLV to confirm your probability inputs are real before you trust the sizing at all.
You don't have to do the arithmetic by hand. Cobia's Tools tab includes a Kelly calculator — enter your estimated probability and the price, and it returns the full-, half-, and quarter-Kelly stakes so you can size a position in seconds without over-betting.
Key takeaways
- Kelly outputs the bankroll fraction to stake: f = (b × p − q) / b, or roughly edge ÷ odds.
- A 55% edge at even money = 10% full Kelly; a better price on the same edge sizes up fast.
- Full Kelly is too volatile and assumes a perfect estimate — sharps bet half or quarter Kelly.
- Kelly is worthless without a real +EV edge; garbage probability in, dangerous stake out.
- Over-betting a wrong edge is the fastest way to ruin — when unsure, bet less. Confirm your edge with CLV.
Cobia ships a Kelly calculator and grades every pick in public.
No deleted losers, no fake units.
Start free at cobiapicks.com
Cobia