The Kelly criterion is a formula that sizes each bet as a fraction of your bank based on your edge: f = (bp – q) / b. Here b is decimal odds minus 1, p is your chance of winning and q is 1 minus p. It can be applied to fixed-odds racing in theory, but it only works if your probabilities are accurate, and that is the hard part.
A worked example in Australian decimal odds
Take a horse at $4.00, so b = 3. The price implies a 25% chance. Say your ratings give it 30%, so q = 0.70. Then f = (3 x 0.30 minus 0.70) / 3 = 0.0667, or 6.67% of your bank. On a $5,000 bank that is about $333. If you rate the horse at 25% or lower, the formula returns zero or a negative number, which means no bet.
Why full Kelly is risky when you are wrong
Kelly assumes your probability is exactly right. Suppose the horse is really a 27% chance. It is still a positive-value bet, but a 6.67% stake is about 2.5 times the true Kelly size of 2.67%. The long-run growth of the bank turns slightly negative at that stake. Theory says staking more than double the correct Kelly fraction shrinks a bank over time. Even a correct 30% horse loses 7 times in 10, so full Kelly also brings sharp drawdowns.
Fractional Kelly
Most people who use Kelly bet a fraction of it, commonly half or a quarter. The table uses the example above on a $5,000 bank. It shows the average growth per bet if your 30% estimate is too high and the true chance is 27%. A smaller fraction trades growth for protection against your own errors. Full Kelly stakes 6.67% of the bank, while quarter Kelly stakes 1.67%.
| Stake | Share of bank | Stake on $5,000 | Growth per bet at true 27% |
|---|---|---|---|
| Full Kelly (30% estimate) | 6.67% | $333 | -0.00114 |
| Half Kelly | 3.33% | $167 | +0.00099 |
| Quarter Kelly | 1.67% | $83 | +0.00090 |
| Full Kelly at true 27% | 2.67% | $133 | +0.00105 |
Source: Calculated from f = (bp – q) / b at $4.00 (b = 3). Growth is expected natural log change in bank per bet. Hypothetical, not a forecast.
Why most punters use flat staking
You need a trustworthy win probability for every runner, and few punters have one. Tote dividends are not known when you bet, so Kelly suits fixed odds far better than exotics. Flat staking is simple, easy to track and never asks you to trust a number you cannot test. See flat staking and percentage staking for the comparison. Nothing here is a promise of profit.
Frequently Asked Questions
What does a negative Kelly result mean?
It means the price is shorter than your probability supports. The formula says stake nothing.
Can you use Kelly on trifectas or quaddies?
Not easily. Tote dividends are set after betting closes, and estimating the chance of an exact finishing order or four winners is far harder than rating one horse.
Does Kelly stop losing runs?
No. It sizes stakes, it does not remove variance. Expect long losing streaks even with a good edge.
Horse racing betting involves financial risk. Bet only what you can afford to lose. For support, visit Gambling Help Online or call 1800 858 858.