The Kelly criterion is the best-known staking formula in betting. It tells you what share of your bankroll to stake on a bet based on two numbers: the odds on offer and your estimate of the true probability. In theory, no other staking method grows a bankroll faster over the long run.

That promise is exactly what makes it dangerous. Kelly is mathematically optimal only when your probabilities are accurate, and in football they almost never are. A small overestimate of your edge can turn the fastest-growing strategy into one that steadily destroys a bankroll.

This guide explains how the formula works, what it gets right, where it breaks down in soccer betting and how experienced bettors use a reduced version of it. It assumes you already have a basic staking framework in place, as outlined in how to manage your bankroll in soccer betting.

 

The Formula in Plain Terms

For decimal odds, the Kelly stake can be written as:

Kelly stake = (probability x odds – 1) / (odds – 1)

The top half of the formula is your edge: how much you expect to gain per unit staked. The bottom half is the net odds: how much you win per unit if the bet lands. So the formula can be read in one sentence: stake your edge divided by the net odds.

If the edge is zero or negative, Kelly stakes nothing. That alone makes it useful as a filter. A bet without a positive expected value has no place in the bankroll, no matter how confident it feels.

A simple example: you estimate that a team has a 50% chance of winning, and the bookmaker offers 2.20.

• Edge: 0.50 x 2.20 – 1 = 0.10, or 10%
• Net odds: 2.20 – 1 = 1.20
• Kelly stake: 0.10 / 1.20 = 0.083, or about 8.3% of the bankroll

For most bettors, 8.3% on a single football match already feels uncomfortable. That instinct is worth listening to, and the reasons become clear further down.

 

How Kelly Sizes Different Types of Bets

Kelly does not simply reward bigger edges with bigger stakes. It also accounts for variance, which depends on the odds. The table below shows how the stake changes across common football prices.

 
The first and last rows show the logic most clearly. The bet at 1.50 has the smallest edge in the table but the largest stake. The bet at 4.00 has the biggest edge but one of the smallest stakes.

That is because a short-priced bet wins often and its outcomes are less volatile, so the formula is willing to commit more. A long-priced bet loses most of the time even when it is good value, so Kelly limits exposure to protect the bankroll during losing runs. The same principle is behind the point made in why a value bet does not mean a high probability of winning: value and win rate are separate things, and staking has to account for both.

 

What Kelly Gets Right

Kelly is built to maximise the long-term growth rate of the bankroll, not the expected profit of any single bet. That is a meaningful difference.

A bettor who always stakes a fixed amount ignores both the size of the edge and the size of the bankroll. A bettor who stakes a large share on every positive bet will eventually hit a losing run that wipes out years of gains. Kelly sits between those approaches. It bets more when the edge is larger, less when variance is higher, and automatically scales stakes down when the bankroll shrinks.

That last feature is often overlooked. Because stakes are a percentage of the current bankroll, a strict Kelly bettor can never lose everything in theory. Each losing bet reduces the next stake, which slows the decline.

For anyone with a genuinely measurable edge, those are real advantages. The problem is the word “genuinely”.

 

The Core Problem: Kelly Assumes You Know the True Probability

The formula treats your probability estimate as fact. In football, every estimate is uncertain, and the uncertainty is usually larger than the edge itself.

Go back to the example at 2.20. You estimated a 50% chance and Kelly suggested 8.3%. Now suppose the real probability was 47%, a difference that is well within the normal error of any football model.

• Real edge: 0.47 x 2.20 – 1 = 0.034, or 3.4%
• Correct Kelly stake: 0.034 / 1.20 = about 2.8%
• Stake actually placed: 8.3%, roughly three times the optimal amount

If the real probability was 45%, the bet had no edge at all, and the correct stake was zero.

This is not a rare scenario. Football matches turn on single events, form is noisy and team news changes late. Bettors also tend to overestimate the edges they find, because bets that look like the biggest value are often the ones where the estimate was most wrong. That is a large part of why finding value bets is much harder than people think.

 

Why Overbetting Is So Costly

The cost of staking too much is not symmetrical. Underbetting slows growth slightly. Overbetting can reverse it completely.

The table below uses the same bet: a true 50% chance at odds of 2.20, where full Kelly is 8.3%. It shows the expected long-term growth per bet at different stake sizes, relative to full Kelly.

 
Two numbers deserve attention. Half Kelly keeps around three quarters of the growth while cutting stakes in half. And at double Kelly, long-term growth falls to zero, even though every single bet still has a positive expected value.

That is the counterintuitive heart of the issue. A bettor can find genuinely profitable bets and still lose money over time purely through stake size. Combined with the estimation problem above, it means that anyone who overestimates their edge by a factor of two or three while using full Kelly is likely operating in the zone where the bankroll does not grow at all.

 

The Drawdowns Are Larger Than Most People Expect

Even with perfectly accurate probabilities, full Kelly produces severe swings. Under the standard mathematical approximation, the chance that a bankroll will at some point fall to a given fraction of its starting level is shown below.

 
A one-in-two chance of seeing the bankroll halved is not a tail risk. It is close to a certainty over a long enough betting career. Psychologically, very few bettors can sit through a 50% drawdown without changing their approach, chasing losses or abandoning a sound method at the worst moment.

The table also explains why fractional Kelly is the norm among professionals. The reduction in growth is moderate, while the reduction in risk is dramatic.

 

Problems Specific to Soccer Betting

Beyond the general weaknesses, football creates several practical issues the basic formula does not handle.

Many bets run at the same time. Kelly assumes one bet is settled before the next is placed. A football weekend can involve ten or more simultaneous positions. Staking full Kelly on each one independently produces a total exposure far higher than the formula intends. A sensible rule is to cap the combined stake on any single round of fixtures.

Bets on the same match are correlated. Backing a team on the Asian handicap and also backing over 2.5 goals in the same game is not two independent bets. If the team underperforms, both can lose together. Treating them as separate Kelly bets understates the risk.

Bookmaker margins shrink real edges. Many apparent edges in football are only a few percent, which is the same size as typical modelling errors. When the edge and the uncertainty are similar in size, the full Kelly stake is mostly a reflection of noise.

Odds move before kickoff. If the price you took drifts away from you after team news, your estimated edge may have been wrong from the start. Measuring whether you consistently beat the final price is one of the few reliable ways to test whether your edges are real, as covered in what closing line value is and why it matters.

Pools and accumulators do not fit the formula well. Pari-mutuel pools pay out according to how many others share the winning result, so the effective odds are unknown when you bet. Multi-leg tickets also compound estimation errors across every selection, which is explained in the math behind soccer pools and why more legs means more risk. Kelly is designed for single bets at known prices, and it should be used there.

 

How Experienced Bettors Use Kelly in Practice

Most serious bettors who use Kelly do not use the formula as written. They use it as a framework with several layers of protection.

1. Use a fraction, not the full stake. Quarter to half Kelly is the most common range. It captures much of the growth while keeping drawdowns at a level most people can tolerate.

2. Shrink your probability towards the market. Instead of using your own estimate directly, blend it with the bookmaker’s no-margin probability. If you estimate 50% and the market’s fair price implies 44%, using something closer to 47% acknowledges that the market contains information you may have missed. This alone removes many of the largest overbets.

3. Set a maximum stake per bet. A hard cap, often in the region of 2% to 3% of the bankroll, prevents a single overconfident estimate from doing serious damage, regardless of what the formula suggests.

4. Cap total exposure per round. Limit the combined stake across all open bets on a matchday, especially when several bets depend on similar assumptions.

5. Recalculate the bankroll on a fixed schedule. Updating stakes after every result is mathematically correct but can encourage emotional decisions. Many bettors reset the bankroll figure weekly or monthly.

6. Track whether the edge is real. If results and closing prices show that estimates are consistently too optimistic, the fraction should be reduced further or the method reviewed.

Applied to the earlier example, a bettor estimating 50% at 2.20 might blend the probability to 47%, which gives a full Kelly stake of about 2.8%, then apply half Kelly for a final stake of around 1.4%. That is roughly one sixth of the original 8.3%, and far more consistent with the real uncertainty in a football match.

 

Kelly Versus Flat Staking

Flat staking, where every bet uses the same amount or the same small percentage, is the main alternative. It is often presented as the beginner’s option and Kelly as the advanced one, but the comparison is more balanced than that.

• Flat staking ignores differences in edge and odds, so it grows more slowly when edges are real. It is, however, very robust to estimation error, because a wrong probability does not change the stake.
• Fractional Kelly grows faster when estimates are good and still limits damage when they are not, as long as the fraction is small and stakes are capped.
• Full Kelly is only appropriate when probabilities are known with real precision, which is rarely the case in football.

For a bettor without a long, verified record of beating the market, flat staking or a very conservative fractional Kelly will usually produce the better real-world outcome. Kelly rewards accuracy, and it punishes overconfidence much harder than flat staking does.

 

A Tool for Precision, Not for Confidence

The Kelly criterion is not dangerous because the mathematics is flawed. It is dangerous because it trusts the input completely, and in football the input is a probability estimate built on incomplete information.

Used at full strength, it magnifies every error in that estimate and produces drawdowns that most bankrolls and most bettors cannot survive. Used as a reduced fraction, combined with stake caps and a realistic view of your own edge, it becomes one of the most sensible staking frameworks available.

The useful question is not whether Kelly works. It is whether your probabilities are good enough to deserve it. For most soccer bettors, the honest answer points to a small fraction of Kelly, and the discipline to keep it there.