How much to put on each bet is a bigger decision than which bet to place. Four plans, what each does to your results, and the one that ruins everyone who tries it.
The same amount on every bet, or one to three units by confidence. It is simple, it makes the record easy to read, and it is what most winning bettors actually use. Its weakness is that it does not adapt: after a good year the unit is too small, after a bad one too large, unless you reset it.
A fixed percentage of the current bankroll, recalculated on a schedule. It compounds gains and cushions losses automatically. Recalculate monthly; recalculating after each bet produces stakes that lurch with variance.
A formula that maximises the long-run growth of a bankroll when you know your edge: stake fraction = (b x p - q) / b, where b is the decimal odds minus one, p is your probability of winning and q is 1 - p.
Because nobody estimates probabilities to the percent, full Kelly overbets in practice and produces violent swings. Half Kelly gives three quarters of the growth with far less volatility; quarter Kelly is what most serious bettors use, which in the example is about 2 percent: the same size as a standard unit. That is not a coincidence.
One, two or three units by conviction. It works if conviction correlates with results; the record will tell you within a few hundred bets. Most bettors find their three-unit bets do no better than their one-unit bets, and go back to flat.
Double the stake after every loss so that the first win recovers everything. At even money it works until a run of losses meets the table limit or the bankroll, and then it loses everything at once. Seven losses in a row at even money happens about once in 128 sequences; a bettor starting at 10 is then asked to stake 1,280 to win 10. Fibonacci, d’Alembert and Labouchere are slower routes to the same place. No staking plan creates an edge; it can only change the shape of the results that an edge, or its absence, produces.
| Plan | Best for | Risk of ruin | Growth |
|---|---|---|---|
| Flat | Almost everyone | Low at 1 to 2 percent units | Linear |
| Percentage | Bettors with a proven record | Low, self-correcting | Compounding |
| Fractional Kelly | Bettors who model probabilities | Low at a quarter, high at full | Optimal for a known edge |
| Martingale | Nobody | Certain, eventually | None |
One to two percent of the bankroll; the basic stake of a flat plan.
Only with honest probabilities and a fraction of the full stake; quarter Kelly is the usual compromise.
Because a long losing run arrives before the bankroll or the limit allows the doubling to continue.