What the Kelly criterion is

The Kelly criterion is a mathematical formula computing the optimal fraction of capital to place on each opportunity, to maximize the long-term growth rate of capital. It rests on two inputs: the win rate and the average win-to-loss ratio.

The formula and its meaning

The Kelly fraction equals the win rate, minus the loss rate divided by the win-to-loss ratio. Written compactly: the optimal fraction equals W minus the quotient of one minus W divided by R, where W is the probability of winning and R is the average win divided by the average loss. If the formula gives a negative number, it means the strategy has no edge and is expected to lose long term; that itself is a signal to stop.

Why it matters and how to apply it

Kelly's deep point is that win rate alone says very little about how much to bet; what matters is the combination of win rate and win-to-loss ratio. However, full Kelly is usually too aggressive in practice because it assumes your estimates are perfectly accurate, which they never are. In reality, most practitioners use a fraction of Kelly, such as a half or a quarter, to sharply reduce volatility and drawdowns, at the cost of some growth rate. Because misestimating your edge usually biases toward the dangerous side, this cautious use is important.

Common mistakes

  • Using full Kelly, betting too large based on an edge estimate that's usually inflated.
  • Forgetting Kelly assumes accurate estimates, while reality always has error.
  • Ignoring the negative-number signal, meaning the strategy has no edge.
  • Treating Kelly as a guaranteed-win formula rather than a sizing framework based on an assumed edge.

FAQ

  • Should I use full Kelly? Most use a fraction of Kelly to reduce volatility and drawdown risk, because full Kelly assumes perfect estimates.
  • What does a negative Kelly mean? It means that by your estimate, the strategy has no edge and is expected to lose long term; it's a sign to reconsider.
  • Does Kelly conflict with fixed-risk rules? Kelly is a sizing method based on edge; beginners should still start with small fixed risk before considering advanced methods.

Checklist

  • ☐ Are my win-rate and win-to-loss estimates based on a long enough sample?
  • ☐ Am I using a cautious fraction of Kelly rather than full Kelly?
  • ☐ If the formula gives a negative number, do I recognize it as a no-edge signal?