Practical guide

Kelly criterion: stake size and risk

The Kelly criterion connects a perceived edge with position size. It is a useful reference but is extremely sensitive to an overstated probability—the most important error in many models and human estimates.

Connect an edge to controlled position size
Connect an edge to controlled position size
01Full Kelly is aggressive
02Fractional Kelly reduces risk
03Limits beat emotion
01

The full-Kelly formula

For decimal odds O and probability p, the bankroll fraction is f = (p × O − 1) / (O − 1). With p = 55% and O = 2.00, full Kelly suggests 10% of bankroll.

If the numerator is zero or negative, there is no mathematical edge and the fraction should be zero. A negative answer is not an instruction to back the opposite outcome.

02

Why full Kelly is often too aggressive

The classical formula assumes that probability is known and opportunities repeat. In reality estimates contain error, positions are correlated and the available price can move. Overstated probability quickly inflates the suggested stake.

Half or quarter Kelly is therefore common in practice. It reduces volatility and the cost of estimation error, but it does not remove the risk of loss.

03

A sensitivity example

At odds of 2.00, a 55% estimate gives 10% full Kelly. If a more cautious estimate is 52%, the suggestion falls to 4%. Only three percentage points changed the position by more than half.

This is why probability calibration matters alongside average accuracy. A stake-sizing formula cannot repair a weak forecasting model.

04

Practical constraints

Set a maximum bankroll fraction in advance, account for correlated positions and never use money required for essential expenses. Update bankroll after settlement, not after an emotional reaction to a streak.

Any calculator is a scenario based on its inputs. It cannot know the next outcome, the quality of the probability source or your tolerance for drawdown.

FAQ

Quick answers

What if Kelly produces a negative number?

Treat the stake as zero: the inputs do not indicate a positive edge.

What is half Kelly?

It is half the full-Kelly fraction, reducing volatility and sensitivity to estimation error.

Does Kelly prevent losing streaks?

No. Long drawdowns remain possible, and an incorrect probability can produce systematic losses.

18+ All materials are informational and do not guarantee any outcome. Predictions are produced by an automated model; the method is described on the Method page. Responsible gambling