Free calculator
Kelly Criterion Calculator
American1.909 decimal
Your own estimate of how often this bet wins, as a percent.
If your number is 2 points too high
At 53.0%, the stake would be $6.50 (0.65% of bankroll).
Half Kelly stakes $27.50, 2.75% of your bankroll.
Before you size a bet with Kelly
- Kelly is only as good as the probability you give it. An estimate a couple of points too high can multiply the stake or turn a pass into a bet, and the row in the calculator shows it for your numbers.
- Full Kelly swings hard. In the standard theory a full Kelly bettor has about an even chance of seeing the bankroll halve at some point along the way. Half or quarter Kelly gives up a little growth for much smaller swings.
- Flat staking is what Agatha uses: every pick is one unit, edge and lean alike, so the record reads as a plain count. The methodology explains why.
How to use it
- Enter the odds you can get, in any format.
- Enter your own win probability for the bet. It is the number that matters most, and the one most likely to be wrong.
- Enter your bankroll and pick a fraction; half or quarter Kelly is the usual choice.
- Read the stake, then check the row that assumes your number is two points too high.
The formula
p is your win probability and d the decimal odds.
b = d - 1 edge = p x d - 1 full Kelly = edge / b (zero when the edge is zero or less) stake = bankroll x full Kelly x fraction growth per bet = p x ln(1 + f x b) + (1 - p) x ln(1 - f)
-110 at 55%: d = 1.9091, b = 0.9091, edge = 55% x 1.9091 - 1 = 5.00%. Full Kelly is 5.00% / 0.9091 = 5.50% of the bankroll; half Kelly is 2.75%, which is $27.50 of $1,000.00. If the true chance is 53%, half Kelly is only 0.65%, or $6.50: two points of probability cut the stake by most of its size.
Edge cases
- No edge, no bet: at or under the price's break-even probability, Kelly's answer is zero.
- Long prices: a small edge on a long price gives a small Kelly fraction, because the stake is lost most of the time.
- Several bets at once: this calculator sizes one bet at a time. Bets that run together call for a smaller total.
- A very large answer usually means the probability is too high. Kelly never stakes the whole bankroll unless the win is certain.
From Agatha's record
Agatha's confidence against the price
Each graded pick carries a confidence, Agatha's own win probability. Here is how often each confidence band won, beside the probability its prices already implied.
| Confidence | Picks | Won | Price implied |
|---|---|---|---|
| under 50 | 28 | 35.7% | 42.7% |
| 50-54 | 1,196 | 51.6% | 53.6% |
| 55-59 | 532 | 57.1% | 57.6% |
| 60-64 | 354 | 64.4% | 63.0% |
| 65-69 | 198 | 63.1% | 66.4% |
| 70-74 | 130 | 71.5% | 71.8% |
| 75+ | 192 | 81.3% | 80.8% |
Across the bands with 50 or more picks, the win rate and the probability the price already implied sit 1.5 points apart on average. The market's own price says nearly what the confidence says.
Over all 2,626 decided picks the win rate was 58.3% against a 59.4% break-even, and on that gap Kelly says do not bet. Agatha stakes every pick at one unit regardless.
As of . See the full track record
Kelly calculator questions
What is the Kelly criterion?
A formula for the share of a bankroll to bet that grows it fastest over many bets, if your win probability is right: edge divided by the net odds. -110 at 55% gives 5.50%.
Why use half or quarter Kelly?
Because the win probability is an estimate. Betting a fraction of Kelly gives up a little growth when you are right and loses far less when you are wrong, and it makes the swings much smaller.
Where does the win probability come from?
From you: a model of your own, or a fair price from a market you trust. The no-vig calculator gives a market's fair chances; Kelly on that same book's price always says pass, because the margin is still in the price.
Does Agatha use Kelly?
No. Every Agatha pick is one unit, flat, so nothing is sized up for conviction. The methodology explains it, and the track record counts every pick the same.
What if Kelly says to bet a large share of my bankroll?
Check the probability first. Real betting edges are small, so a large Kelly stake almost always means the estimate is too high. The losing streaks guide shows how deep normal swings run.