Advanced Staking: Comparing Kelly Criterion vs. Fractional Kelly

In our Basic Staking Guide, we recommended a flat 1-2% stake. That advice is right for almost everyone, and if you take one thing from this page, let it be that.

For advanced bettors who want to maximise growth rate, there is a more aggressive mathematical approach: the Kelly Criterion. This guide explains what it is, why nobody sane uses it at full strength, and how the fractional version actually gets used in practice.

TL;DR: The Kelly Criterion calculates the bankroll fraction that maximises long-term growth: it bets more when your edge is big and less when it's small. Full Kelly is far too violent for real betting — probability estimates are never accurate enough, and the drawdowns are savage. Professionals use a fraction of it (a quarter or an eighth), which keeps most of the growth for a small part of the pain. If you don't have a reliable way of estimating probabilities, Kelly has nothing to work with: stick to flat stakes.

What is the Kelly Criterion?

The Kelly Criterion is a formula used to calculate the optimal bet size to maximise the growth of your bankroll over the long term.

It bets more when the edge is high, and less when the edge is small. That single property is what separates it from flat staking — it's the only mainstream staking method that listens to how good the bet is.

The formula

f = (bp - q) / b

f = fraction of bankroll to bet
b = decimal odds - 1 (net odds)
p = probability of winning
q = probability of losing (1 - p)

Example

Odds: 3.00 (2/1) → b = 2
True probability: 40% → p = 0.40
Edge: large (fair odds would be 2.50)

f = (2 × 0.40 - 0.60) / 2
f = (0.80 - 0.60) / 2
f = 0.10 (10%)

Kelly suggests betting 10% of your bankroll.

Notice what the formula needs to work: a genuine probability estimate. That p = 0.40 has to come from somewhere — a model, a fair-odds line, something better than a hunch. On vibeodds, the EV% figure and the fair odds behind it are exactly that input: a 120% EV bet at 3.00 implies the model rates the horse's chance at 40%, and Kelly turns that edge into a stake size. No edge estimate, no Kelly. It is a tool for people who already know what their edge is, not a way of discovering one.

The danger of Full Kelly

Betting 10% on a single horse is suicide. Why?

1. Model error. Kelly assumes your probability is exactly right. It never is. And the punishment is asymmetric: if your true edge is smaller than you think, Kelly doesn't just grow slower — over-betting an overestimated edge actively shrinks the bankroll. Betting double the true Kelly fraction has an expected growth rate of zero, and beyond that it's negative. Since every bettor overestimates their edge at least some of the time, everyone using Full Kelly on estimated probabilities is over-betting some of the time.
2. Variance. Even when the numbers are right, the ride is horrific. Full Kelly staking routinely produces drawdowns of half the bankroll — a coin-flip's chance of being down 50% at some point is built into the method. A losing streak at 10% stakes compounds fast.
3. You, at 3am, during the drawdown. The maths assumes you keep betting the formula through a 40% downswing. Almost nobody does. A staking plan you'll abandon under pressure is worse than a weaker plan you'll actually follow — the same argument that runs through managing variance.

For these reasons, almost no professional bets Full Kelly.

The solution: Fractional Kelly

Pros use a fraction of the Kelly suggestion — usually Half (50%), Quarter (25%) or Eighth (12.5%).

Using the example above (10% Full Kelly):

Half Kelly: bet 5%
Quarter Kelly: bet 2.5%
Eighth Kelly: bet 1.25%

The trade is heavily in your favour, and this is the part worth remembering: growth falls slowly as you scale down, but risk falls fast. Half Kelly keeps roughly three-quarters of the theoretical growth rate for about half the variance. Quarter Kelly still captures a meaningful share of the growth with drawdowns most humans can actually sit through. The last scraps of growth rate between Quarter and Full Kelly are bought at an absurd price in volatility.

There's a bonus: betting a fraction of Kelly is also insurance against model error. If your probability estimates are systematically a bit too confident (they are), Quarter Kelly on your estimated edge may be close to the correct Kelly on your true edge. The fraction absorbs your overconfidence.

Why Fractional Kelly beats flat staking (in theory)

It sizes to the edge. A 130% EV bet deserves more money than a 104% EV bet. Flat staking treats them identically.
It compounds properly. Stakes scale with the bankroll automatically, up and down.
It can't go bankrupt on any single bet. Stakes are always a fraction of what remains.

And why flat staking often wins in practice: it needs no probability estimate, it's immune to your model's bad days, and you'll actually stick to it. That last one decides more betting careers than any formula.

Practical wrinkles the textbook leaves out

Simultaneous bets. Kelly is derived for one bet at a time. Bet five races in an afternoon at full formula size and your combined exposure is far beyond what the maths intended. Either divide stakes across concurrent bets or run a small enough fraction that the overlap doesn't matter — one more argument for Quarter Kelly or below.

Prices move after you bet. Your edge estimate changes but your stake is already committed (see early price vs SP for the timing game). Fractional staking keeps any single mistimed bet small.

Each-way and lay bets. Kelly extends to both, but the arithmetic changes: for lay betting the relevant quantity is liability, not stake. Don't reuse the win-bet formula unmodified.

Minimum edge filter. Kelly happily bets tiny fractions on tiny edges. Below some EV threshold the edge is more likely model noise than signal, and the correct stake is zero. Most practitioners bet nothing under roughly 103-105% EV regardless of what the formula says.

Conclusion

Unless you have a genuinely calibrated probability source and the temperament of a filing cabinet, stick to flat staking or a conservative Fractional Kelly — a quarter at most.

The goal is to stay in the game.

To find the edges that make Kelly staking possible, visit the Live Odds page and Value view, then track how those high-edge bets perform over time in your Selections view.

Frequently asked questions

Do I need Kelly at all as a casual bettor?
No. Flat 1-2% staking captures most of the benefit with none of the machinery. Kelly starts to matter when you have a measured edge and enough volume for growth-rate differences to compound.

What's the difference between Kelly and proportional staking?
Proportional staking bets a fixed percentage of the current bankroll regardless of edge. Kelly varies the percentage with the edge. Proportional is Kelly with the edge-sensitivity removed — simpler, more robust, less optimal.

What fraction should I use?
If you must have a number: a quarter. If your probability source is unproven, an eighth, which behaves almost like flat staking with a mild tilt toward better bets.

Can Kelly tell me whether a bet is worth making?
Only indirectly — a negative Kelly fraction means no bet. But Kelly assumes the probability you feed it is true. The real "is this worth betting?" question is answered by the expected value, not the staking formula.

Why do I keep hearing Kelly is used by professional gamblers and investors?
Because the underlying insight — size positions to edge and bankroll, never risk ruin — is genuinely how professionals think. Almost all of them apply it fractionally, for exactly the reasons above.