Kelly Criterion, explained.
The Kelly criterion is a position-sizing rule that maximizes expected logarithmic wealth in a specified model of repeated uncertain payoffs.
Why it happens
An edge does not make every stake sensible. Larger stakes increase gains when you win and the damage when you lose. Logarithmic growth puts special weight on preserving the resources needed to participate in later rounds.
The Kelly criterion chooses an allocation that maximizes expected logarithmic wealth in a specified repeated-bet model. It links the size of a position to both its edge and its risk.
Look for this pattern
Compare arithmetic expected gain with growth along a repeated path. The stake with the highest average one-round payoff need not produce the strongest long-run compound growth.
A worked example
An illustrative repeated bet
Suppose a fair even-money payoff is offered on an event with a correctly known 60% win chance.
The simple binary Kelly formula gives 2p − 1 = 20% as its model optimum, not an instruction for a real wager.
Changing the probability, payoff, or constraints changes the result; uncertain estimates make the optimum especially sensitive.
Where this idea is useful
An advantage can be destroyed by betting too much. Position size is part of the decision.
A common misconception
“Kelly guarantees that wealth increases.”
It optimizes a particular expected growth objective. Losing runs and substantial drawdowns remain possible.
What this explanation leaves out
- The model relies on probability estimates and payoff assumptions. Estimation error, constraints, and large drawdowns complicate practical use; this entry is educational.
Why does a probability estimate matter so much?
The calculated edge comes directly from that estimate. If the estimated win chance overstates the true chance, a supposedly optimal stake can be too large or belong to a game with no edge at all.
Which matters more in your model: knowing the edge precisely, or choosing a stake from an imprecise estimate?
Associated thinkers
Associations marked provisional are awaiting source review.