Kelly Criterion Calculator
The Kelly criterion gives the share of your bankroll to stake on a bet with an edge: f* = p − (1 − p) ÷ b, where p is the win probability and b the net odds. For trading, use the win rate and the average win ÷ average loss ratio. The calculator adds fractional Kelly, the amount at risk, the growth curve and what happens if your win rate is a few points lower than you think.
Inputs Example values The defaults are made up. Your win probability and odds are estimates, and the result is only as good as they are. Use a dot for decimals.
- Net odds b (profit per 1 staked)
- –
- Expected profit per 1 staked = p × b − (1 − p)
- –
- Break-even win probability = 1 ÷ (1 + b)
- –
- Full Kelly f* = p − (1 − p) ÷ b
- –
- Your stake = multiplier × f*
- –
- Amount staked
- –
- Growth per bet g(f), your stake · full Kelly
- –
- Your growth as a share of the maximum
- Bets to double, typical path ≈ ln 2 ÷ g
- –
- Zero-growth point fc (g = 0)
- –
Growth curve g(f)
Expected log growth per bet for every stake from 0 to beyond twice Kelly. Hover or tap the chart to read values.
If your win probability is lower than you think
| True win probability | Kelly f* at that probability | Growth at your stake | Growth at the full Kelly you computed |
|---|
Same odds, same stake; only the true probability changes. A negative growth rate means the bankroll shrinks over many bets even though each bet looked positive.
Everything runs in your browser. Nothing is saved or sent.
How to use the Kelly criterion calculator
- Pick a mode. “Betting odds” for a bet that pays fixed odds. “Trading” for a strategy described by its win rate and its average win ÷ average loss.
- Enter your win probability (or win rate). This is an estimate. The calculator cannot check it.
- Enter the payoff. Odds in decimal, American or net form, or the win/loss ratio R for trading.
- Choose a Kelly fraction. Full Kelly maximises long-run growth only if your inputs are exactly right. Half or quarter Kelly gives up some growth for much smaller swings.
- Optional: enter your bankroll to see the amount to stake or risk.
- Read the curve and the table below it. They show how close your stake is to the zero-growth point, and what a 2 or 5 point error in the win probability does.
Kelly criterion formula
John Kelly, a Bell Labs scientist, published the idea in 1956: bet a fixed fraction of current capital so that the long-run growth rate of the capital is as high as possible. He showed that a gambler who does this will, with probability one, “eventually get ahead and stay ahead” of one who uses any other fixed fraction in an endless series of bets. Edward Thorp later gave the form used here for uneven payoffs: “f* = (bp − q)/b”, where q = 1 − p.
- p is the probability the bet wins. In trading mode it is the win rate W.
- b is the net profit per 1 staked when the bet wins. Decimal odds 2.10 mean b = 1.10. In trading mode b is R: a loss costs the amount risked, a win pays R times that amount.
- g(f) is the expected logarithm of the growth factor per bet. Over n bets the typical (median) bankroll multiplies by roughly en × g. It is a long-run average, not a forecast of any single result.
Converting odds to b
| Odds format | Rule | Example | b |
|---|---|---|---|
| Decimal | b = decimal odds − 1 | 2.10 | 1.1 |
| American, positive | b = odds ÷ 100 | +150 | 1.5 |
| American, negative | b = 100 ÷ |odds| | −110 | 0.9091 |
| Fractional | b = the fraction itself | 5/2 | 2.5 |
| Trading | b = R = average win ÷ average loss | 300 ÷ 150 | 2 |
What the growth curve shows
The curve rises from zero, peaks at full Kelly, then falls. It crosses zero at fc, which is often close to twice Kelly. Thorp writes that “often 2f* ≐ fc”. Beyond that point the bankroll shrinks in the long run, even though every bet has a positive expected profit. Thorp's even-money example: with p = 0.53, f* = 6%, and “if the fraction wagered is more than about 12%” the bankroll is driven toward zero.
The curve is flat near the top and steep on the right. Betting less than Kelly costs a little growth. Betting more costs growth and adds risk. In Thorp's words, “overbetting is much more severely penalized than using too small an f* and underbetting.”
Fractional Kelly: growth kept versus stake size
For the calculator's default example (decimal odds 2.10, p = 55%). The last column is the continuous-time approximation 2c − c², the share of maximum growth kept at c times Kelly.
| Multiplier c | Stake | Growth per bet | Share of maximum growth | Approximation 2c − c² |
|---|---|---|---|---|
| 0.25 | 3.52% | 0.477% | 43.7% | 43.8% |
| 0.5 | 7.05% | 0.818% | 75.0% | 75.0% |
| 0.75 | 10.57% | 1.022% | 93.7% | 93.8% |
| 1 | 14.09% | 1.091% | 100.0% | 100.0% |
| 1.5 | 21.14% | 0.812% | 74.5% | 75.0% |
| 2 | 28.18% | −0.045% | −4.1% | 0.0% |
Half Kelly halves the stake but keeps about three quarters of the growth. That is the trade-off behind fractional Kelly. MacLean, Thorp and Ziemba put it this way: “growth is traded off for security with the use of fractional Kelly strategies”.
Why errors in p make full Kelly risky
The formula assumes you know p exactly. You never do. You estimate it from a model, a betting line or a trade history, and the estimate is noisy. Kelly is sensitive to that error because the edge is usually small next to p itself.
- At decimal odds 2.10, f* falls from 14.09% at p = 55% to 10.27% at 53% and 4.55% at 50%. At 47.62% (the break-even point) it is zero.
- If you stake the full Kelly computed from 55% but the true probability is 50%, the growth rate is −0.389% per bet: the bankroll shrinks over time.
- Half Kelly from the same estimate still grows at 0.079% per bet when p is really 50%.
Thorp's advice, from the section “The case for fractional Kelly”: when the edge is an uncertain estimate, “it is wise to assume” the true edge is smaller and to bet below the computed Kelly “by enough to prevent g ⩽ 0”. MacLean, Thorp and Ziemba note “the extreme sensitivity of E log calculations to errors in mean estimates” and say that, to be safe, “the size of the wagers should be reduced.”
A trade history gives a rough estimate. With 100 trades and a 45% win rate, the standard error of the win rate is about √(0.45 × 0.55 ÷ 100) ≈ 5.0 percentage points. A 5 point error is well inside two standard errors. The Kelly criterion spreadsheet computes W and R from a trade log and shows Kelly at the win rate minus two standard errors.
Worked examples
All numbers below are fictional, chosen for the arithmetic, except Example 1, which is Thorp's.
1. Even-money coin with p = 0.53 (Thorp, Example 2.1)
| Step | Arithmetic | Result |
|---|---|---|
| Net odds | even money | b = 1 |
| Full Kelly | 0.53 − 0.47 ÷ 1 | 6.00% |
| Growth per bet | 0.53 × ln(1.06) + 0.47 × ln(0.94) | 0.001801 (0.180%) |
| Bets to double | ln 2 ÷ 0.001801 | ≈ 385 |
| Zero-growth point | 0.53 × ln(1 + f) + 0.47 × ln(1 − f) = 0 | 11.97% |
Thorp reports f* = .06, g(f*) = .001801 and about 385 bets to double. He gives fc as .11973; solving the same equation here gives 0.11971. Both round to “about 12%”.
2. Sports bet at decimal odds 2.10 with a 55% estimate
| Step | Arithmetic | Result |
|---|---|---|
| Net odds | 2.10 − 1 | b = 1.1 |
| Break-even probability | 1 ÷ (1 + 1.10) | 47.62% |
| Full Kelly | 0.55 − 0.45 ÷ 1.10 | 14.09% |
| Half Kelly | 0.5 × 14.09% | 7.05% |
| Stake on a 1,000 bankroll | 7.05% × 1,000 | 70.45 |
| Growth per bet | half Kelly · full Kelly | 0.818% · 1.091% |
| Zero-growth point | g(f) = 0 | 27.91% (1.98 × f*) |
3. No edge: American odds −110 with a 52% estimate
−110 means a 110 stake wins 100, so b = 100 ÷ 110 = 0.9091. f* = 0.52 − 0.48 ÷ 0.9091 = −0.80%. The Kelly fraction is negative, so the calculator says do not bet. The break-even probability is 1 ÷ (1 + b) = 52.38%; a 52% estimate is below it.
4. Trading: 45% win rate, average win twice the average loss
| Step | Arithmetic | Result |
|---|---|---|
| Full Kelly | 0.45 − 0.55 ÷ 2 | 17.50% |
| Quarter Kelly | 0.25 × 17.50% | 4.38% |
| Amount at risk on a 25,000 account | 4.38% × 25,000 | 1,093.75 |
| Growth per trade | quarter Kelly · full Kelly | 1.314% · 2.924% |
| If the true win rate is 43% | 0.43 − 0.57 ÷ 2 | 14.50% |
| If the true win rate is 40% | 0.40 − 0.60 ÷ 2 | 10.00% |
The Kelly stake in trading is the amount you lose if the trade hits its average loss, not the position size. Turn that risk amount into shares or units with the position size calculator.
Using Kelly for trading: the limits
- Fixed win and loss sizes. The trading form treats every win as +R and every loss as −1 times the amount risked. Real trades vary, gaps can make a loss bigger than planned, and fees reduce R.
- One bet at a time. The formula sizes one bet on the whole bankroll. Several open positions that move together are closer to one large bet. MacLean, Thorp and Ziemba list among the bad properties that “the bets may be a large fraction of current wealth”.
- Independent, unchanging odds. Kelly assumes each bet has the same known edge and is independent of the last. Market conditions change; a backtest win rate may not hold.
- Large swings. Even with correct inputs, full Kelly produces deep drawdowns. Check the losing streaks your stake can survive with the risk of ruin calculator.
- Win rate alone says nothing. A 70% win rate with R = 0.3 has no edge: f* = 0.7 − 0.3 ÷ 0.3 = −0.3. See why win rate alone doesn't matter.
FAQ
What is the Kelly criterion formula?
For a bet that pays b per 1 staked when it wins, with win probability p, the Kelly fraction is f* = p − (1 − p) ÷ b. It is the share of the bankroll that maximises the expected logarithm of wealth, which is the long-run growth rate. At decimal odds 2.10 (b = 1.10) and p = 55%, f* = 0.55 − 0.45 ÷ 1.10 = 14.09%.
What does a negative Kelly percentage mean?
It means there is no edge at the odds offered: the win probability is at or below the break-even probability 1 ÷ (1 + b), so every stake has a negative expected profit. The Kelly answer is to bet nothing. At American odds of −110 the break-even probability is 52.38%, so a 52% estimate gives a negative Kelly fraction.
Why do people use half Kelly or quarter Kelly?
Because the inputs are estimates. If the true win probability is lower than estimated, full Kelly overbets, and a large enough overbet makes the long-run growth rate negative. A fraction of Kelly cuts the stake and the swings sharply while keeping much of the growth: half Kelly keeps about 75% of the maximum growth rate in the continuous approximation. Thorp recommends betting below the computed Kelly when the edge is uncertain.
How do you use the Kelly criterion for trading?
Use the trading form f* = W − (1 − W) ÷ R, where W is the win rate and R the average win divided by the average loss. The result is the share of the account to lose if a trade loses, the risk per trade, not the position size. With W = 45% and R = 2, f* = 0.45 − 0.55 ÷ 2 = 17.5%. The formula assumes fixed win and loss sizes and independent trades, which real trading does not meet, so most users stake a fraction of it.
What happens if you bet more than the Kelly criterion?
Growth falls and risk rises. The growth rate per bet drops on both sides of full Kelly, but faster on the overbetting side, and it reaches zero at a stake often close to twice Kelly. Beyond that point the bankroll shrinks over many bets even though each bet has a positive expected profit.
Can you lose money betting the Kelly criterion?
Yes. Kelly maximises the long-run growth rate only when the win probability and payoff are known exactly and bets are repeated many times. Any single bet can lose, losing streaks happen at every stake size, and wrong inputs give a wrong stake. The calculator shows what the formula implies for the numbers entered, nothing more.
Related tools and templates
- Risk of ruin calculator
Chance of losing a set share of the account at a given risk per trade. - Kelly criterion spreadsheet
Excel and Google Sheets: Kelly from a trade log, with a growth table and chart. - Position size calculator
Turn a risk amount and a stop distance into shares or units. - Why win rate alone doesn't matter
Win rate and payoff together decide whether there is an edge.
Sources
- J. L. Kelly, Jr., “A New Interpretation of Information Rate”, Bell System Technical Journal, vol. 35, no. 4, July 1956, pp. 917–926 (DOI).
- Edward O. Thorp, “The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market”, chapter 9 in Handbook of Asset and Liability Management, Volume 1, eds. S. A. Zenios and W. T. Ziemba, Elsevier, 2006: sections 2, 6 and 7.3.
- Leonard C. MacLean, Edward O. Thorp and William T. Ziemba, “Good and Bad Properties of the Kelly Criterion”, in The Kelly Capital Growth Investment Criterion: Theory and Practice, World Scientific, 2011, pp. 563–572.
Web sources retrieved 8 Nov 2026.
Educational tool only, not financial, investment or betting advice. Results come from the numbers you enter and the formulas above; they are not predictions. Win probabilities and payoffs are estimates, and an estimate that is too high makes the Kelly stake too large. Betting and leveraged trading can lose money quickly. All example numbers are fictional except the cited Thorp example.