The Kelly Criterion for Portfolio Allocation

The Kelly Criterion is a mathematical framework for deciding how much capital to allocate to an investment when the goal is to maximize long-term wealth growth.

Originally developed for gambling and information theory, the Kelly Criterion has become an important concept in investing because it addresses a question that traditional valuation methods often leave unanswered:

Even if an investment is attractive, how much of your portfolio should you put into it?

A great investment can still produce a terrible outcome if the position is too large. Conversely, allocating too little capital to your best opportunities can significantly reduce long-term returns.

The Kelly Criterion attempts to find the balance between those two extremes.

At its core, Kelly seeks the portfolio allocation that maximizes the expected logarithmic growth of wealth. In practical terms, it tries to maximize the long-term compound growth rate of a portfolio while accounting for the possibility of losses.

That makes Kelly fundamentally different from simply maximizing expected returns.

What Is the Kelly Criterion?

The Kelly Criterion was developed by John L. Kelly Jr. at Bell Labs in the 1950s. Kelly originally applied the concept to information theory and betting, but the mathematics translates naturally to investment decisions.

In its simplest form, the Kelly Criterion determines what percentage of your available capital should be placed on a favorable bet.

For a binary outcome, the standard Kelly formula is:

f^* = \frac{bp-q}{b}

Where:

$f^*$ = optimal fraction of capital to invest
$b$ = net profit received for each dollar invested if successful
$p$ = probability of success
$q$ = probability of failure, or $1-p$

For example, suppose an investment has:

A 60% probability of gaining 50%
A 40% probability of losing 25%

The Kelly framework weighs both the probabilities and the magnitude of the possible outcomes to determine the allocation that maximizes long-term compounded wealth.

This is an important distinction. Kelly is not asking which investment has the highest expected return. It is asking how aggressively you should allocate capital given the relationship between expected return and risk.

Why Kelly Focuses on Compound Growth

Investors do not experience returns additively. Wealth compounds multiplicatively.

Suppose you begin with $100.

If you gain 50%, your portfolio increases to $150.

If you then lose 50%, your portfolio falls to $75.

Your average arithmetic return was:

\frac{50\% + (-50\%)}{2} = 0\%

Yet you lost 25% of your money.

This happens because losses have a disproportionate effect on compound growth.

A 50% loss requires a 100% gain just to recover.

A 75% loss requires a 300% gain.

A 90% loss requires a 900% gain.

Kelly addresses this problem by maximizing expected logarithmic wealth:

\max E[\ln(W)]

Where $W$ represents future wealth.

Using logarithmic wealth naturally penalizes outcomes that produce large drawdowns or ruin. As a portfolio approaches zero, the logarithm of wealth approaches negative infinity.

This is why Kelly generally avoids allocations that create excessive risk of catastrophic loss.

Expected Return Alone Is Not Enough

Suppose you have two investments.

Investment A

Expected annual return: 15%
Volatility: 15%

Investment B

Expected annual return: 25%
Volatility: 50%

Investment B has the higher expected return, but that does not necessarily mean you should allocate more money to it.

Its much greater volatility creates a significant drag on compound growth.

A useful approximation for geometric growth is:

g \approx \mu - \frac{\sigma^2}{2}

Where:

$g$ = expected compound growth rate
$\mu$ = expected arithmetic return
$\sigma$ = volatility

This relationship helps explain why Kelly cares about both return and variance.

Increasing expected return helps long-term growth.

Increasing variance hurts it.

The optimal portfolio must balance the two.

The Kelly Criterion for an Investment Portfolio

For investments with approximately normally distributed returns, a simplified Kelly allocation for a single risky asset is:

f^* = \frac{\mu-r_f}{\sigma^2}

Where:

$f^*$ = Kelly allocation to the risky asset
$\mu$ = expected return of the risky asset
$r_f$ = risk-free rate
$\sigma^2$ = variance of the asset's return

The numerator represents the investment's expected excess return.

The denominator represents its risk.

This formula reveals one of the most important ideas behind Kelly:

Position size should increase with expected excess return and decrease with risk.

An investment with twice the expected excess return should generally receive a larger allocation.

An investment with twice the volatility should receive a much smaller allocation because volatility enters the equation as variance.

Example: Kelly Allocation to a Single Stock

Assume a stock has:

Expected annual return: 12%
Risk-free rate: 4%
Expected annual volatility: 25%

The expected excess return is:

12\%-4\%=8\%

Variance is:

0.25^2=0.0625

The Kelly allocation is therefore:

f^*=\frac{0.08}{0.0625}=1.28

The theoretical full-Kelly allocation is:

128\%

That means the mathematical solution would allocate 128% of the portfolio to the stock, requiring leverage.

For example:

$100,000 of investor capital
$128,000 invested in the stock
$28,000 borrowed

That does not mean an investor should automatically use that allocation.

The calculation depends heavily on the accuracy of the expected return and volatility assumptions. Small errors can produce dramatically different recommended allocations.

This is one reason many investors use fractional Kelly instead.

Fractional Kelly

Full Kelly maximizes expected long-term logarithmic growth under the assumptions used in the model.

The problem is that investors rarely know the true expected return, volatility, correlations, or probability distribution of an investment.

For that reason, practitioners often use a fraction of the theoretical Kelly allocation.

Common approaches include:

Half Kelly: 50% of the full-Kelly position
Quarter Kelly: 25%
Three-quarter Kelly: 75%

In the previous example, the full-Kelly position was 128%.

A half-Kelly allocation would be:

128\%\times 0.50=64\%

A quarter-Kelly allocation would be:

128\%\times0.25=32\%

Fractional Kelly sacrifices some theoretical long-term return in exchange for lower volatility, smaller drawdowns, and greater protection against estimation errors.

This tradeoff can be extremely attractive in real-world investing.

Why Half Kelly Is So Interesting

One of the useful properties of Kelly sizing is that reducing the allocation below full Kelly can reduce risk considerably while giving up less expected compound growth than investors might assume.

The exact relationship depends on the assumptions, but the broader principle is important:

You do not have to invest at full Kelly to capture much of the benefit of Kelly-style position sizing.

Because expected returns are so difficult to estimate precisely, using half Kelly or even quarter Kelly can create a meaningful margin of safety.

Kelly Allocation Across Multiple Assets

Portfolio allocation becomes more interesting when several investments are involved.

You cannot simply calculate Kelly independently for every stock and combine the results because the assets may be correlated.

Two investments with identical expected returns and volatility can deserve very different portfolio weights if one provides diversification and the other duplicates risks already present elsewhere in the portfolio.

For multiple assets, the approximate Kelly solution is:

\mathbf{w}^* = \Sigma^{-1}(\boldsymbol{\mu}-r_f\mathbf{1})

Where:

$\mathbf{w}^*$ = vector of optimal portfolio weights
$\Sigma$ = covariance matrix of asset returns
$\Sigma^{-1}$ = inverse covariance matrix
$\boldsymbol{\mu}$ = vector of expected asset returns
$r_f$ = risk-free rate
$\mathbf{1}$ = vector of ones

This may look complicated, but the concept is straightforward.

Kelly considers three things:

The expected return of each investment
The risk of each investment
How each investment behaves relative to the others

That third factor is critical.

Correlation Changes the Optimal Allocation

Imagine two stocks that each have:

Expected return: 12%
Volatility: 20%

If their returns are almost perfectly correlated, owning both provides little diversification.

If their correlation is close to zero, combining them can reduce portfolio volatility substantially.

If they are negatively correlated, the diversification benefit can be even greater.

As a result, Kelly may recommend larger combined exposure to two uncorrelated investments than it would to two highly correlated investments.

This leads to an important insight:

Kelly position sizing is a portfolio problem, not merely a stock-by-stock problem.

The attractiveness of an investment depends partly on what else you already own.

Example of Kelly Portfolio Allocation

Consider a simplified portfolio containing two risky investments.

Stock A

Expected return: 12%
Volatility: 20%

Stock B

Expected return: 10%
Volatility: 15%

Assume:

Risk-free rate: 4%
Correlation between A and B: 0.30

The expected excess returns are:

\mu_A-r_f=8\%

\mu_B-r_f=6\%

The variances are:

\sigma_A^2=0.20^2=0.0400

\sigma_B^2=0.15^2=0.0225

The covariance is:

Cov(A,B)=0.30\times0.20\times0.15=0.009

The covariance matrix is therefore:

\Sigma= \begin{bmatrix} 0.0400 & 0.0090\\ 0.0090 & 0.0225 \end{bmatrix}

The Kelly portfolio is found by calculating:

\mathbf{w}^*=\Sigma^{-1} \begin{bmatrix} 0.08\\ 0.06 \end{bmatrix}

The approximate result is:

w_A\approx1.49

w_B\approx2.07

That implies theoretical full-Kelly allocations of approximately:

Stock A: 149%
Stock B: 207%
Total risky exposure: 356%

The remaining portfolio weight would effectively be financed through borrowing.

At first glance, this can look absurdly aggressive. But it illustrates something important about mathematical Kelly portfolios.

When estimated Sharpe ratios are attractive and correlations are relatively low, the unconstrained Kelly solution often recommends substantial leverage.

Real investors therefore commonly impose constraints such as:

No leverage
Maximum 10% or 20% position sizes
Maximum exposure by industry
Maximum total equity exposure
Minimum cash reserves
Fractional Kelly sizing

If the investor used quarter Kelly instead, the approximate allocations would become:

Stock A: 37%
Stock B: 52%
Total risky assets: 89%
Cash or risk-free assets: 11%

That looks much more like a portfolio an investor might actually hold.

Kelly and the Sharpe Ratio

The Kelly Criterion has a close relationship with the Sharpe ratio.

For a single risky asset:

f^*=\frac{\mu-r_f}{\sigma^2}

The Sharpe ratio is:

S=\frac{\mu-r_f}{\sigma}

Therefore:

f^*=\frac{S}{\sigma}

This shows that Kelly favors investments with high expected excess returns relative to their volatility.

More broadly, under the standard continuous-return assumptions, the maximum expected log-growth available from an optimally leveraged risky portfolio is closely related to the square of its Sharpe ratio.

This helps explain why diversification matters so much under Kelly.

If diversification improves the portfolio's expected return per unit of risk, Kelly may justify greater overall exposure while still improving expected compound growth.

The Biggest Problem With Kelly: Estimating Expected Returns

The Kelly mathematics is relatively straightforward.

Estimating the inputs is not.

Volatility and correlations can at least be estimated from historical data, although they can change substantially over time.

Expected returns are much harder.

Suppose your Kelly calculation assumes that a stock will return 15%, but its true expected return is only 8%.

The resulting position may be far too large.

This problem is especially serious because Kelly sizing responds aggressively to estimated excess return.

An investor who overestimates expected returns across a portfolio can end up with excessive concentration or leverage.

For fundamental investors, expected return might therefore be estimated using building blocks such as:

Expected\ Return \approx Earnings\ Growth + Shareholder\ Yield + Valuation\ Change

Rather than assigning a single precise expected return, it may be better to model multiple scenarios.

For example:

Bear case: 5%
Base case: 11%
Bull case: 17%

The investor can then examine how the Kelly allocation changes under each assumption.

If a position only looks attractive under optimistic assumptions, that should reduce confidence in a large allocation.

Kelly and Valuation-Based Investing

Kelly is especially useful when combined with fundamental analysis.

Valuation can help answer:

What return might this investment produce?

Kelly can then help answer:

Given that expected return and the uncertainty surrounding it, how large should the position be?

Suppose you estimate that a stock can produce a 13% annual return over the next decade through:

6% growth in per-share earnings
3% shareholder yield
4% annualized benefit from multiple expansion

If the risk-free rate is 4%, the estimated excess return is 9%.

That expected return can become an input into a Kelly framework.

However, the 4% contribution from multiple expansion is much less certain than a contractual bond yield. A prudent investor should account for that uncertainty.

This is another argument for fractional Kelly and conservative return estimates.

Kelly Does Not Eliminate Risk

Kelly is sometimes misunderstood as a formula that prevents large losses.

It does not.

A full-Kelly portfolio can be extremely volatile.

It can experience substantial drawdowns even when the assumptions are correct.

Kelly's objective is not:

Minimizing volatility
Minimizing maximum drawdown
Preserving capital over every short period
Producing the highest Sharpe ratio
Avoiding losses

Its objective is specifically to maximize expected long-term logarithmic growth.

That objective may be inappropriate for an investor who has:

A short investment horizon
Significant withdrawal needs
Low tolerance for drawdowns
Concentrated outside financial risks
Limited access to liquidity
Large tax consequences from rebalancing
Strict portfolio constraints

The mathematically optimal growth portfolio is not necessarily the psychologically or financially appropriate portfolio for a particular investor.

Overbetting Is More Dangerous Than Underbetting

One of the most valuable lessons from Kelly has little to do with calculating an exact percentage.

It is the asymmetry between betting too much and betting too little.

If the optimal allocation is 20% and you invest 10%, you sacrifice some potential compound growth.

If the optimal allocation is 20% and you invest 50%, you can dramatically increase volatility and reduce long-term compound growth.

At sufficiently high allocations, an otherwise attractive investment can produce negative expected logarithmic growth.

In other words:

Having an investment edge does not justify unlimited position size.

This is particularly important when leverage is involved.

Once an investor moves beyond the Kelly-optimal allocation, additional leverage actually reduces expected long-term compound growth, even though it continues to increase expected arithmetic return.

Practical Ways to Use Kelly in Portfolio Management

Kelly is often more useful as a framework than as a rigid rule.

A practical process might look like this:

Estimate expected returns conservatively. Use realistic assumptions for growth, dividends, buybacks, valuation changes, and other return drivers.
Estimate risk. Historical volatility can provide a starting point, but also consider business risk, leverage, cyclicality, liquidity, and potential permanent capital impairment.
Measure correlations. Position sizes should reflect how holdings interact with one another, not just their standalone characteristics.
Calculate theoretical Kelly allocations. Treat these as reference points rather than automatic portfolio weights.
Apply fractional Kelly. Half Kelly or quarter Kelly may be more appropriate when inputs are uncertain.
Add portfolio constraints. Limit individual positions, leverage, sector exposure, and other concentrations.
Stress-test your assumptions. Recalculate allocations using lower expected returns, higher volatility, and higher correlations.
Rebalance thoughtfully. Kelly weights change as valuations, expected returns, volatility, correlations, and interest rates change.

A Useful Conservative Version of Kelly

For investors who want to incorporate uncertainty explicitly, one practical approach is to reduce the expected excess return before calculating Kelly.

Instead of:

f^*=\frac{\mu-r_f}{\sigma^2}

you might use:

f^*=\frac{\mu-r_f-M}{\sigma^2}

Where $M$ represents a margin of safety for estimation error.

Alternatively, calculate full Kelly using conservative assumptions and then multiply by a fractional-Kelly factor:

f_{practical}=k\times f^*

Where:

0<k<1

For example, if:

f^*=40\%

and you use half Kelly:

f_{practical}=0.50\times40\%=20\%

This creates two layers of protection:

Conservative assumptions
Conservative position sizing

For long-term investors dealing with inherently uncertain forecasts, that may be much more robust than blindly following full Kelly.

Kelly Criterion vs. Traditional Portfolio Optimization

Kelly has similarities to mean-variance portfolio optimization, but the objectives are different.

Traditional mean-variance optimization generally attempts to maximize expected return for a particular level of volatility, or minimize volatility for a desired return.

Kelly specifically maximizes long-term compound growth.

Under certain mathematical assumptions, the two approaches are closely related. Both reward expected returns, penalize variance, and account for correlations.

But Kelly provides something particularly useful: an answer to the question of how much total risk to take.

A traditional efficient frontier can identify the best combination of risky assets without necessarily specifying how aggressively the investor should leverage or deleverage that portfolio.

Kelly provides a theoretical answer by identifying the level of exposure that maximizes expected log growth.

When Kelly Is Most Useful

The Kelly Criterion is particularly useful when:

You have a repeatable investment process
You can make reasonable estimates of expected returns
You are comparing several attractive opportunities
Position sizing materially affects portfolio outcomes
You want to account for correlations between investments
You are deciding whether leverage is justified
You want a mathematical framework for balancing conviction and risk

It is less reliable when:

Outcomes have extreme or poorly understood tail risks
Expected returns are highly speculative
Historical volatility understates true economic risk
Correlations are unstable
Investments are illiquid
Losses can exceed modeled assumptions
The portfolio contains options or other strongly nonlinear payoffs that are modeled incorrectly

In those situations, a simplistic Kelly calculation can create false precision.

The Most Important Lesson From the Kelly Criterion

The greatest value of the Kelly Criterion may not be the exact allocation it produces.

Its most important insight is that investment selection and position sizing cannot be separated.

An investment can have a positive expected return and still deserve only a small position.

Another investment can have a modest expected return but deserve a meaningful allocation because its risk is low and its correlation with the rest of the portfolio is favorable.

Kelly forces investors to think about four variables together:

Expected return
Risk
Correlation
Position size

It also highlights one of the fundamental realities of investing: maximizing long-term wealth is not the same thing as maximizing expected return.

Compound growth depends on surviving losses, controlling volatility, and avoiding allocations large enough to destroy the benefit of an investment edge.

For most investors, full Kelly is probably better viewed as a theoretical upper benchmark than as a literal portfolio prescription. Fractional Kelly, conservative expected-return assumptions, diversification, and sensible position limits can make the framework much more practical.

Used this way, the Kelly Criterion becomes less of a betting formula and more of a disciplined way to answer one of the hardest questions in portfolio management:

Given the expected return, the risk, and everything else I already own, how much of my money should I actually put into this investment?