The Kelly Criterion is the mathematical answer to “how much should I bet” on a positive-expected-value wager. It tells you the stake size that maximises long-term bankroll growth given the edge you have and the odds you're getting. The calculator below handles the maths. The rest of this article covers what the formula means, when to use it, and why almost nobody serious uses full Kelly.
The Kelly Criterion formula
Kelly fraction = edge / (decimal odds − 1)
Edge is your expected-value percentage as a decimal. 4% edge becomes 0.04. The output is the fraction of your bankroll to stake on this specific bet.
Worked example 1: full Kelly on a +EV bet
You find an AFL H2H bet at Bet365 with Collingwood at $2.10. You estimate the true probability of Collingwood winning at 52%.
Edge calculation: EV = (0.52 × 2.10) − 1 = 0.092, or 9.2%.
Kelly fraction: 0.092 / (2.10 − 1) = 0.092 / 1.10 = 8.36% of bankroll.
On a $10,000 bankroll, full Kelly says bet $836 on this game. That's a lot of money on a single sports bet. It's also exactly what the maths says to do if you're absolutely confident in your 52% probability estimate. The word “absolutely” is doing important work here, and we'll come back to it.
Worked example 2: Kelly on a smaller edge
A different bet: NRL H2H, Penrith at $1.80, true probability estimated at 58%.
Edge: EV = (0.58 × 1.80) − 1 = 0.044, or 4.4%.
Kelly fraction: 0.044 / (1.80 − 1) = 0.044 / 0.80 = 5.5% of bankroll.
On a $10,000 bankroll, full Kelly says bet $550. Smaller than the Collingwood bet despite the lower odds, because the edge is also smaller. Kelly scales naturally with both edge size and odds — the calculator above handles both inputs.
Why full Kelly is a trap
Full Kelly has two serious problems in practice.
Estimation error. Kelly's optimality depends on your edge estimate being exactly right. Not approximately — exactly. In sports betting, your true-probability estimates always have error bars. You think the true probability is 52%. It might actually be 50%. If you stake full Kelly based on a 52% estimate and the reality is 50%, you're betting 8% of bankroll on a coin flip. Your expected geometric growth rate becomes slightly negative. Full Kelly under over-estimation doesn't just reduce your growth — it inverts it.
Variance. Even if your estimates were perfect, full Kelly sizing produces bankroll swings most people can't tolerate emotionally. Drawdowns of 50% are normal under full Kelly. A 50% drawdown requires a 100% recovery just to break even. Most bettors quit or tinker with their strategy before the recovery happens, destroying the compounding that was supposed to work in their favour.
Fractional Kelly: the practical solution
The fix is to bet a fraction of the Kelly stake. The calculator above supports the four common options:
- Full Kelly (100% of Kelly fraction): mathematically optimal, practically brutal. Use only if you're absolutely confident in your edge estimates (nobody should be).
- Half Kelly (50%): still aggressive. Variance is significantly lower than full Kelly but drawdowns are still meaningful. Requires serious emotional discipline.
- Quarter Kelly (25%): the most common sizing for serious AU advantage bettors. Balances long-term growth against variance. Tolerant of moderate estimation error. Recommended starting point.
- Eighth Kelly (12.5%): very conservative. Growth rate is lower but drawdowns are manageable. Appropriate during periods when you're uncertain about your edge, or during emotional recovery from previous drawdowns. See the eighth-Kelly drawdown piece for why this matters in practice.
On the Collingwood example above (full Kelly = 8.36%):
- Quarter Kelly: 2.09% of bankroll = $209 on $10,000 bankroll
- Eighth Kelly: 1.04% of bankroll = $104 on $10,000 bankroll
Both are real bets but not life-ruining if they lose. Over hundreds of similar placements with genuine +EV, expected growth is meaningful. The full bankroll guidecovers the detailed case for quarter Kelly as the default.
When Kelly doesn't work
Kelly assumes several things that don't always hold in practice:
Accurate edge estimation. Kelly is only as good as your edge estimate. Estimates off by even 1-2 percentage points can flip the formula from correct to actively harmful. The fix is fractional Kelly, which tolerates estimation error.
Independent bets. Kelly assumes each bet is independent. If you're placing correlated bets (multiple legs of the same match, or multiple AFL games that share public-money dynamics), sizing each at Kelly produces effective total exposure larger than Kelly intended. Scale down when betting correlated positions.
Infinite time horizon. Kelly maximises long-term geometric growth assuming you keep betting indefinitely. If you have a defined end-point (a specific bankroll target, a particular account nearing restriction), Kelly may not be the optimal size. More conservative sizing tends to be appropriate when time horizons are finite.
Kelly for multi-bet scenarios
Kelly applied to multi bets (parlays) produces very small recommended stakes because multis compound variance dramatically. This is the maths telling you something useful: multis aren't a good use of bankroll even when each individual leg is +EV. See the multi bets piecefor the full argument.
Kelly on AU same-game multis (SGMs) typically returns a negative number, indicating you shouldn't bet at all. This matches the reality that SGM pricing at AU bookmakers is almost universally −EV after accounting for correlation assumptions baked into the line prices.
Tracking Kelly performance
If you're using Kelly sizing consistently, two things to track:
Realised growth vs expected growth. Your actual bankroll growth rate should approximate the Kelly-predicted growth rate over a sufficient sample (1,000+ bets). Significant divergence means either your edge estimates are wrong or your Kelly sizing implementation has a bug.
Closing line value. Average CLV should roughly match average claimed edge. If you're claiming +4% EV on your bets but your average CLV is +1%, your true edge is probably closer to +1%, and your Kelly sizing is too aggressive. See the CLV guide for the detailed relationship.
Frequently asked questions
What Kelly fraction should I use for sports betting?
Quarter Kelly (25% of the full Kelly stake) is a reasonable default for most serious AU advantage bettors. Start there and adjust based on your tolerance for variance and confidence in your edge estimates.
How do I calculate Kelly for a bet?
Use the formula: Kelly fraction = edge / (decimal odds − 1). The calculator above handles it automatically — input your bankroll, the odds, your estimated true probability, and the Kelly fraction you want to apply.
Does Kelly Criterion work for betting on favourites?
Kelly works for any +EV bet regardless of whether the bet is on a favourite or underdog. The formula doesn't care about odds direction — it cares about edge size and payout structure. A +4% EV bet at $1.50 odds gets a smaller Kelly stake than a +4% EV bet at $3.00 odds because the payout ratio is smaller.
Is Kelly sizing the same as flat staking?
No. Flat staking uses the same dollar amount on every bet regardless of edge or odds. Kelly scales the stake based on edge size and odds. Flat staking is robust and easy but leaves meaningful growth on the table; Kelly extracts more edge but requires accurate inputs.
Do professional bettors use Kelly?
Most professional sports bettors use fractional Kelly or a related scaling approach. Full Kelly is rare even among pros because the variance is too painful and the estimation error in edge estimates makes fractional sizing safer.
Daniel writes about the maths underneath advantage betting — expected value, Kelly sizing, closing line value, bankroll theory. Translates the theoretical side into practical decisions AU punters can actually apply.