Flat Betting or Kelly: Which Strategy Should You Choose?

Once a bettor has defined their bankroll, one essential question arises:

how much should you stake on each bet?

Two approaches regularly appear in professional bankroll management:

Flat Betting

and:

the Kelly Criterion

Flat Betting is based on a simple principle: stake the same Unit on every selection.

The Kelly Criterion follows a different philosophy: the Stake varies according to the estimated Edge on each bet.

At first glance, Kelly may seem more sophisticated and therefore more efficient.

But the reality is more complex.

A Stake Sizing method is only effective if the information used to determine the Stakes is sufficiently reliable.

In this article, we will compare both approaches to understand their advantages, limitations and the situations in which each may be appropriate.

As in the previous articles in this series, we will use the following reference:

Initial bankroll: €5,000

1U = €50

which represents:

1% of the initial bankroll

What Is Flat Betting?

Flat Betting consists of staking exactly the same Unit on every bet.

In our example:

Bankroll: €5,000

1U = €50

Every bet is therefore placed with:

€50

It does not matter whether a bet appears particularly attractive or only slightly profitable.

For example:

The amount risked always remains the same.

This simplicity is one of the main strengths of Flat Betting.

What Is the Kelly Criterion?

The Kelly Criterion works differently.

It attempts to adjust the Stake according to the estimated Edge.

Its classic formula is:

f = (bp - q) / b*

where:

f* = fraction of the bankroll to stake

b = decimal odds - 1

p = estimated probability of winning

q = 1 - p

The greater the estimated Edge, the larger the Stake Kelly may recommend.

Conversely, when the Edge becomes smaller, the Stake decreases.

If no positive Edge is identified:

Kelly recommends no bet.

The Fundamental Difference Between Flat Betting and Kelly

The difference is not simply about the size of the Stakes.

It is mainly about how the two methods treat the available information.

Flat Betting essentially says:

“I believe my selections are good enough to bet, but I do not claim to know the exact Edge of each one with sufficient precision.”

Kelly says something different:

“I can estimate the true probability of each bet accurately enough to adjust my exposure according to its Edge.”

This distinction is fundamental.

Kelly requires much more information.

And, above all:

much more precise information.

A Simple Example With a €5,000 Bankroll

Suppose we identify five bets.

With Flat Betting:

1U = €50

We obtain:

Total exposure:

€250

or:

5% of the bankroll

With Kelly, the Stakes might instead become:

Total exposure:

€500

or:

10% of the bankroll

Kelly concentrates more capital on the opportunities considered the strongest.

That can be highly valuable.

But only if the ranking of the Edges is correct.

Flat Betting Does Not Try to Optimize Every Bet

This is an essential point.

Flat Betting does not claim to determine the mathematically optimal Stake for each selection.

Its objective is different.

It primarily seeks:

simplicity

discipline

comparability

risk control

and:

a cleaner measurement of selection quality

Flat Betting therefore deliberately accepts that it may not fully exploit potential differences in Edge between bets.

In exchange, it significantly reduces dependence on the precision of those estimates.

Kelly Attempts to Exploit Differences in Edge

Kelly follows a different logic.

Not all Value Bets are necessarily equal.

Consider two bets offered at odds of:

2.00

For the first, you estimate the true probability at:

51%

For the second:

55%

The first has a theoretical EV of:

+2%

The second:

+10%

With Flat Betting, you might stake:

€50 on each

Kelly, however, considers it logical to allocate more capital to the second bet.

Mathematically, this logic is coherent.

But it depends entirely on one question:

can you really distinguish a 51% bet from a 55% bet?

The Real Problem Is Not Calculating Kelly

The Kelly formula itself is relatively simple.

The real challenge is determining:

p

meaning:

the true probability of the event.

Consider odds of:

2.00

If you estimate:

p = 51%

Full Kelly recommends:

2%

If you estimate:

p = 53%

Kelly recommends:

6%

If you estimate:

p = 55%

Kelly recommends:

10%

A difference of only four percentage points therefore transforms a Stake of:

2%

into:

10%

The difficulty is not mathematical.

It is statistical.

Flat Betting Is More Robust to Estimation Errors

Suppose you believe a bet has an exceptional Edge.

Your model estimates the probability of winning at:

55%

at odds of:

2.00

Full Kelly therefore recommends:

10% of the bankroll

or:

€500

But imagine that the true probability is actually only:

51%

Your Edge still exists.

However, the theoretically appropriate Kelly is no longer 10%.

It is:

2%

or:

€100

Your estimation error has caused you to stake five times more than the true Edge would justify.

With our 1U Flat Betting approach, you would simply have staked:

€50

Flat Betting therefore sacrifices some potential optimization in exchange for greater robustness to estimation errors.

Kelly Rewards Precision

It would, however, be incorrect to conclude that Kelly is simply more dangerous.

Kelly becomes particularly interesting when the probabilities being used are:

reliable

calibrated

reproducible

and:

validated over a sufficiently large Sample Size

Suppose a quantitative model can produce properly calibrated probabilities.

In that case, treating:

a 1% Edge

and:

an 8% Edge

exactly the same may leave part of the advantage unexploited.

Kelly can then allocate more capital to situations where the estimated Edge is greater.

The Importance of Probability Calibration

To use Kelly seriously, a model must do more than simply find good bets.

It must also produce good probabilities.

This is a major distinction.

Suppose a model assigns a probability of:

60%

to 1,000 events.

If the model is properly calibrated, approximately:

600 of those events

should occur.

If only:

540

regularly occur, the model is probably overconfident.

Poor Calibration can become particularly dangerous with Kelly.

Why?

Because an overconfident model mechanically produces:

Stakes that are too large.

Flat Betting and Selection Quality

One of the major advantages of Flat Betting is that it allows the quality of selections to be observed more directly.

If every bet is worth:

1U

then the result mainly depends on:

the quality of the selection process

rather than the ability to correctly modify Stake sizes.

This is precisely the logic behind the:

Flat Stakes Summary (1U)

This metric allows strategies to be compared without variations in Stake obscuring their underlying performance.

Selection Edge and Sizing Edge

When analyzing a strategy using variable Stakes, it can be useful to distinguish between two potential sources of performance.

Selection Edge

The ability to identify bets with positive Expected Value.

Sizing Edge

The ability to stake more when the Edge is genuinely greater.

These are two different skills.

A bettor may be excellent at identifying Value Bets but poor at determining which ones have the greatest Edge.

In that case, Flat Betting may work perfectly well while Kelly actually worsens the results.

Conversely, a model capable of accurately estimating the magnitude of the Edge may benefit from variable Stake Sizing.

Example: When Kelly Really Improves the Result

Imagine two groups of bets.

Group A

100 bets

Average true Edge:

+2%

Group B

100 bets

Average true Edge:

+8%

If you use exactly the same Stake on both groups, you exploit both Edges equally.

But if your model can reliably identify this difference, it may be rational to allocate more capital to Group B.

This is precisely the type of situation where Kelly can add value.

The important word, however, remains:

reliably

If the 8% Edge exists only in the Backtest because of Overfitting, increasing the Stake becomes dangerous instead.

The Role of Sample Size

One strategy shows:

ROI +12% after 80 bets

Another shows:

ROI +4% after 5,000 bets

The first may look more impressive.

But can we really conclude that its Edge is three times greater?

Not necessarily.

Sample Size directly affects the uncertainty surrounding the Edge.

And that uncertainty should influence how much confidence is placed in Stake Sizing.

The smaller the sample, the harder it may be to justify an aggressive Kelly approach.

The Problem of Overfitting

Kelly and Overfitting can form a particularly dangerous combination.

Imagine a strategy optimized on historical data.

The Backtest shows:

ROI +10%

The model therefore assumes a significant Edge.

Kelly then recommends large Stakes.

But if part of that historical ROI is the result of Overfitting, the true future Edge may be much smaller.

You then have two simultaneous errors:

an overestimated Edge

and:

an oversized Stake

Flat Betting naturally limits the impact of the second problem.

Flat Betting and Variance

Flat Betting obviously does not eliminate Variance.

With:

1U per bet

you can still experience:

-5U

-10U

or more.

A losing streak remains entirely possible.

However, constant Stake sizes make the distribution of results easier to understand.

Each loss represents:

-1U

Each win depends only on the odds.

This makes statistical analysis of the strategy easier.

Kelly and Variance

Kelly can amplify bankroll fluctuations when Stakes become large.

Imagine the following sequence under Flat Betting:

+1U

-1U

+1U

-1U

With Kelly, the same selections might have received:

1U

4U

1.5U

3U

The financial result could become very different.

Selection Variance is then combined with:

Stake Sizing Variance.

This adds another dimension of risk.

What Is the Impact on Drawdown?

With Flat Betting, Drawdown is generally easier to interpret.

If you lose:

20U

with 1U = 1% of the initial bankroll, that represents:

€1,000

on our reference bankroll.

With Kelly, Drawdown also depends on the Stake sizes assigned to losing bets.

If several bets considered strongly +EV lose in succession, Drawdown can accelerate.

Full Kelly can therefore generate significant Drawdowns even when the strategy genuinely possesses an Edge.

Flat Betting and Risk of Ruin

With a sufficiently small Unit relative to the bankroll, Flat Betting makes Risk of Ruin relatively straightforward to control.

In our example:

Bankroll: €5,000

Stake: €50

1U = 1%

a considerable number of Units would have to be lost to completely exhaust the capital.

That obviously does not mean Risk of Ruin is zero.

A strategy with negative Expected Value will eventually damage the bankroll if used for long enough.

But a small Stake slows this deterioration and provides more time to identify the problem.

Kelly and Risk of Ruin

Within its theoretical framework, Kelly is specifically designed to avoid Overbetting while maximizing the logarithmic growth of capital.

The practical problem, once again, comes from uncertainty.

The optimal Kelly calculated using:

the true probability

is not necessarily the same Kelly obtained using:

your estimate of that probability.

The less accurate the estimate, the further the Stake can move away from the level genuinely justified by the Edge.

Flat Betting Makes Discipline Easier

Flat Betting also provides an important behavioral advantage.

Before starting, the rule is already known:

every bet = 1U

There is therefore no need to make emotional decisions such as:

“This one deserves 2U.”

“I'm very confident, so I'll put 4U on it.”

“I lost the previous two, so this one has to win.”

The Stake is determined independently of emotion.

This helps reduce the risks associated with:

Chasing

Overconfidence

and:

Recency Bias

Kelly Does Not Mean “Bet According to Your Confidence”

This is a common misunderstanding.

Saying:

“I'm very confident, so I'll bet 5U”

is not an application of Kelly.

Kelly requires a quantitative estimate.

For example:

Estimated probability: 54%

Odds: 2.00

Calculated Edge

Calculated Kelly Stake

Without an explicit and sufficiently reliable probability estimate, this is not really a Kelly approach.

It is simply subjective variable Stake Sizing.

Subjective Variable Stakes Can Be the Worst Compromise

There is actually a third situation between Flat Betting and Kelly:

variable Stakes based on intuition.

For example:

1U if the bet looks decent

2U if it looks good

3U if the bettor feels very confident

This approach has some of Kelly's disadvantages without necessarily benefiting from its mathematical rigor.

It increases Stake Sizing Variance while leaving substantial room for behavioral biases.

For variable Stakes to be genuinely rational, they should ideally be based on a measurable and reproducible methodology.

Half Kelly: A Possible Compromise

It is not necessary to choose only between:

Flat Betting

and:

Full Kelly

One intermediate solution is:

Half Kelly

If Full Kelly recommends:

6%

Half Kelly recommends:

3%

If Full Kelly recommends:

2%

Half Kelly recommends:

1%

This approach preserves part of the Edge-based Stake adjustment while reducing exposure.

Quarter Kelly: An Even More Conservative Approach

Quarter Kelly uses:

25% of Full Kelly

If Full Kelly recommends:

8%

Quarter Kelly gives:

2%

With a €5,000 bankroll:

€100

or:

2U

This approach can be particularly interesting when a probabilistic model is available but uncertainty around its estimates remains significant.

Kelly With a Stake Cap

Another approach is to use Kelly while imposing a maximum Stake.

For example:

Maximum Stake = 2% of the bankroll

In our example:

2% = €100 = 2U

If Kelly recommends:

0.8%

you use:

0.8%

If it recommends:

1.5%

you use:

1.5%

If it recommends:

4%

you cap the Stake at:

2%

This preserves part of Kelly's logic while controlling extreme estimates.

Comparing the Different Approaches

Consider a bankroll of:

€5,000

and suppose a bet produces a theoretical Full Kelly of:

8%

We obtain:

The difference in exposure is considerable.

The chosen Stake Sizing system can therefore fundamentally change the risk profile of a strategy.

Which System Provides the Best Comparison Between Tipsters?

When comparing the underlying quality of several tipsters or strategies, Flat Betting has an important advantage.

Suppose:

Tipster A: +20U

Tipster B: +25U

If every bet was calculated at:

1U

the comparison is relatively straightforward.

But if Tipster B uses Stakes ranging from:

1U to 8U

the comparison becomes more complicated.

Part of the difference may come from Stake Sizing rather than the quality of the selections.

This is why the:

Flat Stakes Summary (1U)

is a particularly important performance metric.

Flat Betting Does Not Mean Every Bet Has the Same Quality

This is an important nuance.

Using 1U on every selection does not necessarily mean believing that every bet has exactly the same Edge.

It may simply mean:

“I do not consider my estimate of the differences in Edge precise enough to justify different Stake sizes.”

That can be a perfectly rational approach.

Recognizing uncertainty is an essential part of Risk Management.

Kelly Does Not Necessarily Mean Taking More Risk

Kelly can also recommend less than 1U.

Suppose:

1U = 1%

and Kelly calculates:

0.4%

The Stake would then be:

0.4U

or:

€20

Kelly can therefore reduce exposure when the estimated Edge is small.

The method itself is not inherently aggressive.

It is mainly:

Full Kelly combined with large or overestimated Edges

that can lead to substantial Stakes.

The Role of Odds

Odds also influence the comparison.

A bet at:

1.50

and a bet at:

4.00

do not have the same Variance profile.

Flat Betting may use the same Unit on both.

Kelly indirectly incorporates this difference through its formula.

But once again, the calculation depends on the quality of the probability estimate.

An estimation error at high odds can have significant consequences.

The Role of CLV

CLV can be used as a complementary indicator when assessing the quality of the process.

Suppose a strategy uses Kelly and regularly assigns large Stakes to certain selections.

If those selections systematically generate poor CLV, it may be worth checking whether the model is overestimating its Edge.

Conversely, consistently positive CLV can provide an interesting signal regarding the quality of the prices obtained.

However, CLV does not directly provide:

the exact true probability

and therefore cannot, by itself, be used to calculate Kelly.

Opening Odds and Closing Odds

Analyzing Opening Odds and Closing Odds can also help understand how a strategy behaves.

One strategy may perform well when entering the market very early.

Another may identify Value later.

Comparing entry prices with Closing Odds helps determine whether selections consistently beat the market.

In a Kelly framework, this information can be particularly useful when testing the credibility of the estimated Edge.

The Role of Liquidity

Liquidity can also limit the practical application of Kelly.

Suppose a model recommends:

a 5% Stake

on a large bankroll.

The theoretical amount may exceed what can actually be placed at the observed odds.

The theoretically optimal Stake is then no longer necessarily the Stake that can realistically be executed.

Risk Management must therefore also incorporate:

market depth

and:

actual execution capacity.

Multiple Simultaneous Bets: Flat Betting Is Simpler

Suppose ten bets are available at the same time.

With Flat Betting at:

1U per bet

the maximum exposure is immediately visible:

10U

or:

10% of the initial bankroll

if all bets remain open simultaneously.

With Kelly, all Stakes must be added together.

If the recommendations are:

1% + 2% + 4% + 1.5% + 3% + 2% + 5%...

overall exposure can quickly become significant.

You then need to think not only bet by bet, but also at portfolio level.

Correlation Makes Kelly Even More Complex

Imagine several bets involving:

the same league

the same team

the same type of market

or:

the same model

These bets may be correlated.

Applying Kelly independently to each one can underestimate overall risk.

Flat Betting obviously does not eliminate correlation.

But constant and generally limited Stake sizes make exposure easier to control.

Which Method Is Better for a Beginner?

For someone starting to manage a bankroll seriously, Flat Betting has several educational advantages.

It allows them to learn how to analyze:

ROI

Variance

Drawdown

Sample Size

CLV

and:

Risk of Ruin

without immediately adding the complexity of variable Stake Sizing.

Most importantly, it helps answer a fundamental question:

“Does my selection method actually have an Edge?”

Before trying to optimize Stake sizes, it is logical to verify that the Edge exists.

Which Method Is Better for an Advanced Quantitative Model?

The situation may be different for a quantitative approach capable of producing:

probabilities

uncertainty intervals

verified Calibration

a substantial historical dataset

and:

robust out-of-sample analysis

In this case, Kelly or Fractional Kelly can become much more relevant.

Stake Sizing can then become a logical extension of the probabilistic model.

Flat Betting Can Serve as a Benchmark

Even when a strategy uses Kelly, it can be useful to keep Flat Betting as a reference.

You can compare:

Actual performance with Kelly

with:

Flat Stakes Summary (1U) performance

This helps answer an essential question:

does Stake Sizing actually improve the strategy?

If Kelly increases returns while dramatically worsening Drawdown, the decision is not necessarily obvious.

The relationship between:

Return / Risk

should be analyzed rather than final profit alone.

How Should Flat Betting and Kelly Be Compared Properly?

A serious comparison should ideally use exactly the same selections.

You could then measure:

The question should therefore not simply be:

“Which method makes more money?”

It should instead be:

“Which method produces the most appropriate return and risk profile given the actual quality of my estimates?”

Is a Hybrid Approach Possible?

Yes.

There are many possibilities between strict Flat Betting and Full Kelly.

For example:

Flat Betting at 1U

Kelly capped at 2U

Half Kelly

Quarter Kelly

or:

Fractional Kelly + Stake Cap

It is also possible to use a Stake grid.

For example:

Low Edge = 0.5U

Medium Edge = 1U

High Edge = 1.5U

However, this approach is only rational if the Edge categories are objectively defined and statistically validated.

Otherwise, it risks becoming nothing more than a subjective confidence system.

When Does Flat Betting Make Sense?

Flat Betting can be particularly relevant when:

• an Edge appears to exist but its exact magnitude is difficult to measure;
• probabilities are not sufficiently calibrated;
• the Sample Size remains limited;
• you want to compare strategies cleanly;
• simplicity is a priority;
• you want to limit the impact of Stake Sizing errors;
• you want to reduce emotional biases;
• you first want to evaluate selection quality.

In these situations, the simplicity of Flat Betting is a genuine strength.

When Does Kelly Become Interesting?

Kelly becomes more relevant when:

• reliable probabilities are available;
• their Calibration has been verified;
• the model has a substantial Sample Size;
• the Edge can be measured reproducibly;
• the model has been tested out of sample;
• potential Drawdown is understood;
• correlations are controlled;
• total exposure is monitored;
• Liquidity allows the Stakes to be executed;
• clear risk limits have been established.

Under these conditions, adapting the Stake to the Edge can become rational.

Mistakes to Avoid

Whether using Flat Betting or Kelly, several mistakes remain dangerous.

1. Betting without a demonstrable Edge.
2. Increasing Stakes after a loss.
3. Confusing personal confidence with probability.
4. Overestimating Edge from a small Sample Size.
5. Ignoring Variance.
6. Ignoring Drawdown.
7. Ignoring correlation between bets.
8. Constantly changing the method after a few results.
9. Using Kelly with poorly calibrated probabilities.
10. Using variable Stakes based only on intuition.
11. Evaluating a strategy only by its Profit.
12. Failing to compare results with the Flat Stakes Summary (1U).

Flat Betting or Kelly: The Real Question to Ask

Ultimately, the question is not:

“Which method is more sophisticated?”

Nor even:

“Which method can generate the highest return?”

The real question is:

“How good is the information I have available to determine my Stake?”

If you only know that a selection appears to have positive Expected Value, but cannot accurately measure its Edge:

Flat Betting offers a particularly robust framework.

If you have a model capable of producing reliable, calibrated and validated probabilities:

Kelly can take Stake optimization further.

Between the two, Fractional Kelly and Stake Caps provide intermediate levels of caution.

Conclusion

Flat Betting and Kelly are not simply two competing methods.

They correspond to two different levels of information.

Flat Betting says:

“I have identified an opportunity, but I deliberately limit how much weight I give to my estimate of its Edge.”

Kelly says:

“I consider my estimate of the Edge sufficiently precise to determine my exposure mathematically.”

With our reference bankroll of:

€5,000

and:

1U = €50 = 1%

Flat Betting provides a simple and transparent structure that is particularly useful for analyzing the true quality of selections.

Kelly can go further by adapting the Stake to the Edge.

But that sophistication comes at a price:

it requires a much more precise probabilistic estimate.

The most appropriate Stake Sizing method therefore does not depend solely on the formula being used.

It depends primarily on:

data quality

model reliability

Probability Calibration

Sample Size

Variance

Drawdown

and:

the ability to genuinely measure Edge.

Before trying to optimize every euro invested, a much more fundamental question must therefore be answered:

“Do I really know my Edge accurately enough to justify varying my Stakes?”

If the answer is uncertain, the simplicity of Flat Betting may be an advantage rather than a limitation.

FAQ: Flat Betting or Kelly

What Is Flat Betting?

Flat Betting means using the same Stake on every bet, for example 1U per selection.

What Is the Main Difference Between Flat Betting and Kelly?

Kelly adjusts the Stake according to the estimated probability and Edge, while Flat Betting uses a constant Stake.

Is Kelly More Profitable Than Flat Betting?

Not necessarily. Kelly can better exploit genuine differences in Edge, but it depends heavily on the accuracy of the probabilities being used.

Why Is Flat Betting Simpler?

Because it does not require you to precisely estimate the Edge of every bet in order to determine the Stake.

Can You Use Half Kelly or Quarter Kelly?

Yes. Half Kelly uses 50% of Full Kelly, while Quarter Kelly uses 25%. This reduces exposure to estimation errors.

Can Kelly Be Capped?

Yes. A Stake Cap can, for example, limit each bet to 2% of the bankroll even when Kelly recommends more.

Does Flat Betting Mean Every Bet Has the Same Edge?

No. It simply means using the same Stake, potentially because differences in Edge cannot be measured precisely enough to justify different Stake sizes.

Which Method Is Better for Comparing Tipsters?

The Flat Stakes Summary (1U) is particularly useful because every selection is recalculated using the same Unit.

Can Kelly Be Used Without a Probabilistic Model?

A rigorous application is difficult because Kelly requires an explicit estimate of the true probability.

Which Method Should You Choose?

The choice mainly depends on your ability to estimate and calibrate probabilities. Flat Betting reduces dependence on those estimates, while Kelly allows Stake sizes to be adapted when Edge can be measured with sufficient reliability.