The Kelly criterion answers one narrow question: given a repeated bet with a known probability of winning and a known payoff, what fraction of capital should be staked each time to maximise the growth rate of the capital over many bets? For a simple bet with win probability p, loss probability q and odds b to one, the fraction is p minus q divided by b.
Take a strategy with a 40% hit rate that returns two times risk on winners. Kelly gives 0.40 minus 0.60 divided by 2, which is 0.10. Ten percent of the account per trade. Anyone who has traded knows what a 10% risk per trade feels like in practice, and that instinct is correct: it is why almost nobody trades full Kelly, and why the honest use of the formula is as an upper bound.
What the number is actually maximising
Kelly maximises the expected logarithm of wealth, which in plain terms means it maximises the compound growth rate. It does not maximise expected profit, it does not minimise risk, and it makes no reference at all to how the equity curve behaves in between. A bet size above Kelly lowers growth and raises risk at the same time, which is the one genuinely useful thing the formula tells you: there is a level of aggression beyond which you are being punished twice.
Below Kelly, the trade-off is ordinary. Half the fraction gives roughly three quarters of the growth rate with dramatically less variance. That asymmetry is the reason half Kelly is the standard practitioner setting. You give up a quarter of theoretical growth and remove most of the misery, and it is not a close call.
Full Kelly and the drawdown nobody signs up for
Under full Kelly, a drawdown of 50% is a routine event over a long enough sequence, not a disaster scenario. The mathematics is comfortable with that because the account is never wiped out: sizing shrinks as equity falls, the same fixed fractional mechanism that keeps a 1% rule alive. The trader is not comfortable with it. Nobody keeps executing a system correctly while it is down half.
That is the real objection. Kelly assumes an unemotional operator who will place the next bet at the correct size regardless of the last twenty results. If the operator abandons the system inside the drawdown, the growth rate the formula promised was never available. Sizing has to be set to what will actually be executed, which is a psychological constraint, not a mathematical one. Drawdown tolerance is the binding input.
The inputs are the weak point
Kelly needs p and b. Traders do not have them. They have estimates from a sample of past trades, and those estimates carry error in both directions. Overestimating the edge inflates the fraction fast: a strategy you believe wins 45% of the time at 2R gives a Kelly of 17.5%, and one that really wins 38% gives 7%. A modest error in the hit rate has more than doubled the recommended size.
Three things corrupt the estimate in practice. Small samples, where fifty trades is nowhere near enough to pin down a hit rate. Overfitted backtests, where the parameters were tuned on the same data the statistics come from. And non-stationarity: an edge measured across 2024 volatility may not exist in a different regime. Real trading also has costs Kelly's clean bet ignores, including slippage, spread and swap.
Because errors in the edge estimate push the fraction up more often than down, any Kelly number computed from your own trade history should be treated as an optimistic figure before it is halved, not after.
Continuous outcomes and multiple positions
The classic formula assumes a binary bet: you win b or you lose one. Trading is not binary. Stops slip, targets are taken partially, some trades scratch at breakeven. For a return distribution rather than a coin flip, the general form uses expected return divided by variance, which is where the Kelly fraction and a Sharpe-style ratio start to look like the same idea.
The bigger practical problem is that Kelly sizes one bet at a time. Traders hold several at once, often correlated. Two positions at half Kelly in instruments that move together are close to one position at full Kelly, with none of the diversification the arithmetic silently assumed. Any Kelly-derived number has to be divided across the correlated cluster, not applied to each leg. This is the same leak that breaks the 1% rule, only louder, because the starting fractions are larger.
How to use it without being used by it
Compute Kelly from your own records, not from a strategy's advertised statistics. Take a sample large enough to mean something, strip out the outlier that made the quarter, and use costs as realised rather than theoretical. Then take a fraction of the result: a half is standard, a quarter is common among people managing other people's money, and the difference between them is variance rather than expectation.
Then compare the answer to what your risk rules already permit. If the quarter Kelly number is smaller than your 1% rule, use the smaller one. If it is larger, the fixed rule wins anyway. In practice the formula ends up working as a sanity check that catches oversizing rather than a sizing method in its own right, which is the correct job for it. Prop traders with hard daily limits should treat those limits as the real constraint and use Kelly only to confirm they are not already too aggressive under the firm's drawdown rules.
"Kelly is a ceiling, not a target. Everyone I know who traded full Kelly ended up trading a fraction of it, usually after learning why the hard way."
— Roman Onta, Executive Director, SINGUARD
Key Takeaways
- Kelly maximises long run compound growth for a known edge. It says nothing about the drawdown taken on the way.
- Full Kelly routinely produces drawdowns around half the account, so half or quarter Kelly is the practical setting.
- The formula is only as good as the estimated hit rate and payoff, and estimation errors push the fraction up more often than down.
- Kelly sizes one bet. Correlated positions must share a single Kelly budget rather than each taking the full fraction.
Frequently Asked Questions
What is half Kelly and why is it recommended?
Half Kelly means staking half the fraction the formula returns. It keeps roughly three quarters of the theoretical growth rate while cutting the variance and the depth of drawdowns sharply.
Can I use the Kelly criterion with a small trading account?
You can compute it, but the trade history on a small account is usually too short to estimate the edge reliably, so a fixed percentage risk rule is a safer basis for sizing.
Does Kelly apply if I hold several positions at once?
Not directly. The formula assumes one bet at a time, so correlated open positions have to share one Kelly budget rather than each being sized at the full fraction.
About the Author
Roman Onta is an Executive Director at SINGUARD. He builds the Prop Firm CRM, the Broker CRM, Scalegram and CopySignals side by side with his brother Alex Onta, and he helped on the design of eTrader, the division Alex built and leads. His ground is worldwide payment processing, AML compliance and the corporate structures brokers are built on, work the two of them carry together, shaped by executive roles in the UAE and international corporates. He lives and works in Dubai for most of the year. Meet the executive duo leading Singuard's five divisions.