Kelly Criterion

Kelly Criterion And Bankroll Management

The Kelly Criterion is an increasingly used betting strategy that seeks to maximise profitability by determining stake sized based upon the perceived value edge over the odds on offer. The Kelly Criterion computes the optimal amount to be placed upon a bet, with the larger the edge, the bigger the stake.

The system works on the assumption of a fixed bankroll, from which a percentage for each bet will be derived.  Advantageous odds with overlays are leveraged in your favour with the Kelly Criterion – the approach ensures that you don’t bet so much to jeopardise your bankroll, but you don’t bet too little to stunt its growth.

The Kelly staking plan can be calculated using the following formula:

s = ((p -1) * (q)) – (1-q) / (p-1) x 100

s denotes the % of bank to be staked
p denotes the decimal odds of a selection
q denotes the percentage estimate chance of the selection (must be displayed as a decimal –i.e. 40% would be 0.4)


It can seem fairly complicated at first but is based upon sound logic.

Example. We have a horse available at odds of 3.0 (33.33% chance of winning), but we believe the true odds of success are 2.50 (40% chance of winning).

s = ((3.0 -1) * (0.4)) – (1-0.4) / (3-1) * 100
s = (2 * 0.4) – (0.6) / 2 *100
s = (0.2)/2 *100
s= 10%

The Kelly Staking Plan suggests that we should stake 10% of our betting bank on the selection, based on our edge.

The plan is based upon sound mathematical principles that extenuate profit levels, but staking 10% of a betting bank can be difficult to execute for the more risk averse bettor. Consequently, some adopt the Kelly Criterion, but divide the final stake by 2 or 4.

The Kelly Criterion method of staking is a sound approach, but the issue is defining your edge, as it is simply a matter of judgement. As per the example above, if you believe your selection to be a 40% chance, when in fact it is a 36% chance, you will be over staking. The opposite also applies, and bets can be under staked if the edge is underestimated.