Sizing decides whether you survive the variance
Two people with the same edge can end in very different places depending on stake size. Too large and ordinary losing runs wipe out the bankroll before the edge shows up. Too small and the bankroll is safe but grows slowly. Bet sizing is about surviving uncertainty, not maximising any single result.
Flat staking
Stake the same amount each time — typically 1–2% of the starting bankroll. Flat staking is simple, easy to audit and forgiving of edge estimates that are too optimistic. It is the default for anyone whose probabilities are not rigorously calibrated.
The Kelly criterion
f* = (p × b − q) / b // b = decimal − 1, q = 1 − p
32% at +280: f* = (0.32 × 2.8 − 0.68) / 2.8 = 7.7%
Kelly maximises long-run growth if p is exactly right. It never is. Overestimating p by a few points can more than double the Kelly stake while the true edge is small or negative.
| Fraction of Kelly | Long-run growth vs full | Variance vs full |
|---|---|---|
| Full | 100% | 100% |
| Half | 75% | 25% |
| Quarter | 44% | 6% |
Half Kelly keeps about three-quarters of the growth with a quarter of the variance, and it builds in a safety margin for an overstated edge. That trade is why fractional Kelly is the common choice among people who use Kelly at all.
Kelly assumes independent bets, a known probability, and no simultaneous exposure. Correlated bets on the same game, or several bets at once, need smaller stakes than the single-bet formula suggests. Never size from a probability you cannot defend.