Kelly Criterion Calculator

How to use the Kelly Criterion calculator

1. Enter your honest estimate of the true win probability — not the sportsbook's implied probability. If you think a -110 spread should actually close at -130, your estimated probability is ~56.5%, not 52.4%.
2. Enter the American odds you're being offered right now.
3. Enter your full bankroll (the number you're willing to go to zero with, not your net worth).
4. Choose a Kelly fraction. Full Kelly is too aggressive for almost everyone. Sharp bettors typically use 0.25 (quarter Kelly) because probability estimates are noisy and full Kelly amplifies estimation error into ruinous drawdowns.

The Kelly formula

For a binary bet at decimal odds d with win probability p:

f* = (bp - q) / b

where:
  b = d - 1       (net decimal odds; profit per $1 staked)
  p = win probability
  q = 1 - p       (loss probability)
  f* = fraction of bankroll to stake

If f* is negative, the formula is telling you not to bet — your estimated edge is below the vig. Positive values are the Kelly-optimal stake under the assumption that your probability estimate is exactly right. It almost never is, which is why fractional Kelly exists.

Worked example

The Chiefs are -150 to beat the Raiders on Sunday. Your model (or the sharp-money signal in Delta) tells you the fair line is -200, which corresponds to a 66.7% win probability. Your bankroll is $2,000.

• American -150 = decimal 1.667, so b = 0.667
• p = 0.667, q = 0.333
• f* = (0.667 × 0.667 − 0.333) / 0.667 = 0.166 → 16.6% of bankroll
• Full Kelly stake: $332. Quarter Kelly stake: $83.

Notice how aggressive even a correctly-estimated edge gets at full Kelly. If your probability estimate is off by just a few points, full Kelly can easily suggest staking 30%+ of your roll — and a five-bet cold streak at that size can halve your bankroll. Quarter Kelly gives you most of the long-run growth with a fraction of the variance.

When sharp bettors actually use Kelly

Pros don't run Kelly on gut-feel win probabilities. They use it when they have a calibrated probability model — a bottom-up projection, a no-vig line from sharp books, or a sharp-money signal showing where limits are absorbing real money. Three common applications:

• Sizing off a no-vig fair line. Devig the best market (usually Pinnacle or Circa), compare to the price you can get at a softer book, and stake quarter Kelly on the gap.
• Sizing off a sharp-money signal. When Delta flags a whale trade on an exchange, the implied probability of that bet (sized against depth of book) becomes your p.
• Sizing off a model. If your NFL power ratings say KC -3, and the line is KC -1.5, you have a 1.5-point edge. Convert to probability using a push chart, then Kelly.

Common mistakes

• Using full Kelly. Full Kelly maximizes log-growth but assumes perfect probabilities. Real-world estimates are noisy; fractional Kelly is the practical answer.
• Recomputing bankroll after every bet. This works in theory but amplifies variance after losses (you bet smaller when you're down, so recovering takes longer). Most pros reset bankroll weekly or monthly.
• Kelly on correlated bets. Kelly assumes each bet is independent. Two same-game parlays on the same team are basically one bet — stake as if it's one bet.
• Ignoring bet limits. If Kelly says $1,500 but the book will only take $400, you're not getting Kelly sizing — you're getting max-bet sizing. Shop lines.

Related tools

• All calculators
• Sharp line movement tracker
• Sharp betting tools overview
• Live EV bets

FAQ