Fractional Kelly scales a Kelly fraction calculated from your estimated win probability and average win-to-loss ratio; it does not make uncertain inputs certain. If a small change in the estimated win rate erases the apparent edge, the full-Kelly output is especially sensitive to estimation error. Compare several conservative input scenarios, account for net trade outcomes, and treat the calculator's half- and quarter-Kelly numbers as model outputs rather than drawdown limits.
What the calculator assumes
The Kelly calculator accepts a win rate, average win, average loss, and optional account equity. It converts the win rate into a probability W and divides average win by average loss to get payoff ratio R:
W = win rate percent ÷ 100
R = average win ÷ average loss
Kelly fraction f = W − (1 − W) ÷ R
The result is shown as a percentage. Half Kelly is f ÷ 2; quarter Kelly is f ÷ 4. If equity is entered and the Kelly fraction is positive, the optional dollar output is half Kelly times equity. The calculator does not accept sample size or a confidence interval, and it does not model the order of past outcomes, correlated trades, changing payoff distributions, slippage, or funding. A point estimate from a backtest is still an estimate.
The Stanford paper on risk-constrained Kelly gambling, fetched 2026-11-06, studies an added drawdown-probability constraint alongside the classical growth objective. This illustrates that the standard Kelly fraction and a trader's acceptable drawdown are separate questions. The simple calculator does not implement the paper's risk-constrained method.
Worked example: a modest change cuts the suggested fraction
Suppose a strategy's measured outcomes show a 55% win rate, average winning trade of 1.5R, and average losing trade of 1R. Here, R is just a consistent unit of loss; entering $150 and $100 gives the same ratio as entering 1.5 and 1.
First calculate the payoff ratio and Kelly fraction:
W = 55 ÷ 100 = 0.55
R = 1.5 ÷ 1 = 1.5
f = 0.55 − (1 − 0.55) ÷ 1.5
f = 0.55 − 0.30 = 0.25, or 25%
The calculator therefore shows 12.5% half Kelly and 6.25% quarter Kelly. If equity is $10,000, the code's optional half-Kelly amount is $10,000 × 0.25 ÷ 2 = $1,250. This dollar amount is the output of the formula, not a guaranteed maximum loss or a suggested amount to put at risk on a trade.
Now stress-test a lower win-rate estimate while holding the average win and loss fixed. At 50% wins:
f = 0.50 − (1 − 0.50) ÷ 1.5
f = 0.50 − 0.3333… = 0.1667…, or about 16.67%
half Kelly ≈ 8.33%
At 45% wins:
f = 0.45 − (1 − 0.45) ÷ 1.5
f = 0.45 − 0.3667… = 0.0833…, or about 8.33%
half Kelly ≈ 4.17%
A ten-point change from the initial estimate reduces the half-Kelly output from 12.5% to about 4.17%, with the same payoff ratio. If the win rate fell below the break-even level for that ratio, the calculated Kelly fraction would be negative. The tool can show a negative percentage; its optional risk amount is floored at zero. The example uses hypothetical inputs and does not establish a real strategy's win rate.
Use the calculator to enter your own average outcomes and compare the fractions. Try a range that reflects plausible measurement error instead of selecting the one estimate that produces the largest positive number.
Why a historical win rate can be fragile
A short record can be noisy. A strategy may have few trades, or trades may all come from one market regime. A higher observed win rate in that sample does not prove that future trades will have the same probability. The calculator reports the result for the number entered; it does not assess whether the sample supports that estimate.
Trade definitions can change the result. A backtest may count a position as one trade while a live account records partial exits, re-entries, or multiple fills. Fees, funding, spread, and slippage can convert a small gross winner into a net loser. Measure win rate and average outcomes consistently from net results where possible.
Average payoff also has uncertainty. A fixed win rate with a lower average win or a larger average loss produces a smaller Kelly result. Large outlier winners can pull the average up, while a rare gap or liquidation can create a loss larger than the typical losing trade. Test changes to both sides of the ratio, not just the win rate.
Trades may not be independent. Several crypto positions can share the same market exposure. A string of losses during one regime is not equivalent to unrelated coin flips if trades are correlated. Standard Kelly uses a repeated-outcome model; this calculator does not model dependencies or simulate losing sequences.
What fractional Kelly does and does not change
Half Kelly and quarter Kelly are simple multipliers on the fraction generated by the inputs. If the full fraction is 25%, half Kelly is 12.5% and quarter Kelly is 6.25%. A lower fraction reduces the amount implied by the model, but it cannot correct a win rate or payoff ratio that was measured incorrectly. Nor does choosing a fraction guarantee that losses stay within a preferred drawdown.
For intuition about a losing sequence, imagine a separate simplified model in which each loss really removes a fixed fraction s of the current account, with no gains between losses. After n consecutive losses, remaining equity is equity × (1 − s)^n. At a hypothetical 4% account loss per trade for five consecutive losses, the remaining fraction is 0.96^5 ≈ 81.54%, a drawdown of about 18.46%. At 8%, it is 0.92^5 ≈ 65.91%, a drawdown of about 34.09%. These are arithmetic illustrations, not probabilities that the calculator estimates. A Kelly fraction or its optional dollar output is not automatically equal to the realized loss on a trading position; stop execution, leverage, gaps, and contract rules all matter.
To translate a defined dollar risk and stop distance into a position quantity, use the position size calculator. To examine the payoff from a planned target and stop, use the risk-reward calculator. Those tools address stop-based sizing and reward-to-risk; they do not validate the probability inputs for Kelly.
Common mistakes
- Treating a backtest win rate as a known future probability.
- Entering the win rate without the average win and average loss from the same sample.
- Mixing gross winners with net losers, or using different trade definitions across inputs.
- Assuming half Kelly or quarter Kelly removes estimation error or correlated drawdown risk.
- Treating the optional equity amount as a stop-loss amount or a maximum possible loss.
- Reading a negative Kelly result as a short-trade signal. It only says the supplied assumptions do not yield a positive fraction under this formula.
- Ignoring rare losses, partial fills, fees, funding, or slippage when estimating the average outcomes.
FAQ
Does the calculator account for how many trades produced my win rate?
No. It uses the win-rate percentage you enter and has no trade-count or uncertainty input. You need to evaluate whether your sample is large and representative enough separately.
Why can a small win-rate change move the Kelly result so much?
Kelly depends on both win probability and payoff ratio. When the estimated edge is near zero, a modest input change can reduce the fraction sharply or make it negative.
Does quarter Kelly prevent a large drawdown?
No. It scales the model's fraction down, but actual drawdown depends on outcome sequence, correlation, execution, and whether the model's inputs reflect future results.
Should fees and funding be included in average win and loss?
Use a consistent basis. Net outcomes that include relevant trading costs are usually more useful for estimating the realized payoff distribution than mixing pre-cost and after-cost figures.
Is the half-Kelly dollar output the amount I can lose?
No. It is half the calculated fraction multiplied by entered equity. It is not tied to a stop, liquidation price, or worst-case execution.
This guide is for educational purposes only and is not financial advice.