The Kelly Criterion

What Is The Kelly Criterion?

The Kelly Criterion is a position-sizing method that helps you answer a practical question: how much of your account should you risk when you believe you have an edge?

Rather than trying to predict the market, Kelly focuses on stake size. If your win rate and payoff are strong, it suggests sizing up. If they’re weak — or uncertain — it pushes you to scale down, or step aside.

Used responsibly, the Kelly Criterion can add discipline to your trading plan. Used carelessly, it can encourage over-sizing, especially when your inputs are based on small samples or optimistic assumptions.

Quick Summary

• What it does: estimates a fraction of your capital to allocate when you have a measurable edge.
• What you need: an estimate of your win rate (W) and your win/loss ratio (R).
• How traders apply it in practice: many use fractional Kelly (for example, ½ Kelly or ¼ Kelly) to reduce volatility and limit the impact of estimation errors.
• Important caveat: it’s not a universal “allocate X% to every position” rule — and correlations between trades can materially increase risk.

Want to practise position sizing without putting real money on the line? Open an AvaTrade demo account and test different sizing approaches in real market conditions with virtual funds.

The Kelly Formula And a Worked Example

At its core, the Kelly Criterion calculates an “optimal” fraction of your capital to allocate to a trade based on two inputs:

• W: your probability of winning (win rate)
• R: your average win relative to your average loss (win/loss ratio)

A commonly used version for trading is:

Kelly fraction (f*) = W − (1 − W) / R

Where:

• f* is the fraction of your account the formula suggests allocating (or risking, depending on how you structure your trades)
• W is expressed as a decimal (e.g., 55% = 0.55)
• R is your average win divided by your average loss (e.g., average win £200, average loss £100 → R = 2)

Worked Example

Let’s say your strategy shows:

• W = 0.55 (you win 55% of the time)
• R = 1.5 (your average win is 1.5 times your average loss)

Plugging into the formula:

f* = 0.55 − (0.45 / 1.5)
f* = 0.55 − 0.30
f* = 0.25

In plain terms, full Kelly suggests allocating 25% of capital to that opportunity.

That number often surprises traders — and it highlights an important point: full Kelly can be aggressive, particularly when your inputs (W and R) are not highly reliable. That’s why many traders prefer fractional Kelly to smooth results and reduce drawdown risk.

Fractional Kelly (What Most Traders Use)

Instead of using 25%:

• ½ Kelly would use 12.5%
• ¼ Kelly would use 6.25%

Fractional Kelly doesn’t “break” the concept — it acknowledges that real-world trading includes slippage, changing market conditions, and estimation errors that the clean maths does not fully capture.

To see how sizing affects performance, try running the same strategy with full Kelly, ½ Kelly, and ¼ Kelly on an AvaTrade demo account. Compare volatility and drawdowns before you consider any approach for live trading. Trading involves risk.

How To Estimate W And R Without Guesswork

The Kelly Criterion is only as good as the numbers you feed into it. In trading, the two key inputs are:

• W (win rate): the percentage of trades that finish profitable
• R (win/loss ratio): your average win divided by your average loss

The challenge is that both values can look “excellent” on a small sample — and then fall apart in live markets. That’s why estimation discipline matters as much as the formula itself.

Estimating W (Win Rate)

A practical approach is to calculate win rate from a meaningful sample of trades that reflects how you actually trade (same setup rules, same market conditions, similar execution).

Key considerations:

• Use enough data: a handful of trades is not evidence of an edge. The smaller the sample, the more likely the win rate is distorted by randomness.
• Avoid cherry-picking: include losing streaks, sideways markets, and periods of higher volatility, not only “good weeks”.
• Check stability: if your win rate swings wildly month-to-month, treat W as uncertain and size down accordingly.

Estimating R (Win/Loss Ratio)

R is driven by your strategy’s structure:

• where you take profit,
• where you cut losses,
• and how consistently you follow those rules.

To estimate R:

• calculate your average win across profitable trades,
• calculate your average loss across losing trades,
• then divide average win by average loss.

A few cautions:

• Outliers matter: one unusually large winner can inflate R and make Kelly suggest oversizing. Consider using a “typical” win/loss measure alongside the simple average.
• Execution costs count: spreads, commissions, slippage, and financing can reduce real-world payoffs. If you ignore them, your R is likely overstated.

A Sensible Baseline: Be Conservative First

If you are unsure about W or R, it is usually safer to:

• assume slightly worse numbers than your backtest shows, and
• apply fractional Kelly rather than full Kelly.

This approach sacrifices theoretical “maximum growth” in exchange for a smoother equity curve and a lower chance of a drawdown that you cannot psychologically or financially tolerate.

Before using any sizing model live, test your assumptions. Run your strategy on an AvaTrade demo account, record at least a meaningful run of trades, and calculate W and R from results that include realistic costs and market conditions.

What To Do If Kelly Is Zero or Negative

One of the most useful features of the Kelly Criterion is also one many traders overlook: it can tell you when not to take a trade.

When you calculate the Kelly fraction (f*), you may get:

• f* > 0: the maths suggests a positive edge (in theory, a position size exists)
• f* = 0: you have no edge after accounting for your estimated win rate and payoff
• f* < 0: your edge is negative — the trade is not favourable on the numbers you have

If f* Is Zero

A result near zero typically means one of two things:

• the strategy is close to “break-even” once losses are included, or
• Your estimates of W and R are too uncertain to justify risk.

In practice, this is a signal to reduce size significantly, tighten the rules, or improve the strategy rather than forcing trades.

If f* Is Negative

A negative Kelly fraction is a clear warning: based on your estimated win rate and payoff, you do not have a positive edge. In a disciplined risk framework, the default response is:

• do not take the trade, or
• rework the setup (entry, exit, stop placement, target logic) until the expected payoff improves, and then re-measure.

This point matters because Kelly is not designed to “make a bad strategy work”. It is a sizing framework for strategies with an edge — and it will penalise you when the edge disappears.

Why This Happens More Often Than Traders Expect

Even if your strategy looks profitable on paper, f* can drop to zero or negative when you account for:

• realistic transaction costs (spread, slippage, financing),
• changing volatility,
• regime shifts (a strategy that worked in a trend may struggle in a range),
• or overfitting in backtests.

That is why many traders treat Kelly as a decision filter as well as a sizing tool: if the edge is not clear, the size should not be either.

If your Kelly result is often near zero, use an AvaTrade demo account to gather more data and stress-test your strategy across different market conditions. Focus on improving the edge first — position sizing works best when the underlying setup is genuinely robust.

The Assumptions Behind Kelly (And Why They Matter In Real Trading)

The Kelly Criterion is elegant mathematics — but it comes with assumptions that don’t always hold in live markets.

Understanding these assumptions is essential because the most significant mistakes with Kelly tend to happen when traders treat the output as a certainty rather than an estimate.

1) Your Probabilities Are Accurate

Kelly assumes your win rate (W) and win/loss ratio (R) are measured accurately and will remain relevant going forward.

In practice:

• your sample may be too small,
• costs may be understated,
• and market behaviour can change.

If W and R are wrong, Kelly can recommend a position size that is far too large.

2) Market Conditions Are Stable

Kelly works best when the “game” doesn’t change. Trading is rarely like that.

Strategies often perform differently across regimes, such as:

• trend vs range,
• low vs high volatility,
• normal vs news-driven conditions.

If your strategy’s edge is regime-dependent, your Kelly number should be treated as dynamic, not fixed.

3) Outcomes Are Independent (Or Close Enough)

Kelly also assumes that each bet/trade is largely independent. In real portfolios, trades can be linked in ways that aren’t obvious:

• multiple positions may be exposed to the same macro theme,
• assets can move together during risk-on/risk-off shifts,
• and correlations can rise sharply in stressed markets.

This matters because correlated losses can cluster — making drawdowns deeper than the Kelly framework expects.

4) The Payoff Profile Is Well-Behaved

Kelly-style sizing tends to assume payoffs are reasonably consistent. But trading returns can be skewed by:

• occasional large gaps,
• sudden liquidity drops,
• and “tail events” that overwhelm typical stop-loss behaviour.

This is another reason many traders prefer fractional Kelly even when the maths suggests a higher allocation.

A Practical Checklist Before Using Kelly

Before applying Kelly sizing to any strategy, ask:

• Do I have enough data to estimate W and R with confidence?
• Have I included real-world costs and slippage?
• Does the strategy behave differently across market regimes?
• Are my trades correlated or exposed to the same risk factors?
• Can I tolerate the drawdowns that full Kelly can generate?

If any answer is “not sure”, the conservative response is to size down and prioritise robustness over theoretical growth.

If you want to explore sizing approaches safely, start by applying fractional Kelly on an AvaTrade demo account and track drawdowns across different market conditions. Only consider live deployment once results are stable and risk is clearly defined.

Kelly Vs Fixed-Percentage Risk (Which Is More Practical?)

Many traders size positions using a fixed percentage rule — for example, risking 1% of account equity per trade. It’s simple, consistent, and easy to execute.

Kelly is different. It tries to adapt position size to the strength of the edge. In theory, that can accelerate growth when conditions are favourable. In practice, it can also increase volatility and drawdowns if your estimates are wrong.

Fixed Percentage Risk

How it works: you risk the same fraction of equity on every trade (e.g., 1%), regardless of the setup.

Typical advantages:

• straightforward to apply across markets and timeframes,
• stable risk exposure from trade to trade,
• less sensitive to estimation error (you don’t need a precise W and R).

Common limitation: it doesn’t distinguish between “high-confidence” and “low-confidence” trades — all setups get the same risk budget.

Full Kelly

How it works: you size according to the calculated Kelly fraction (f*).

Typical advantages:

• mathematically optimised for growth under ideal assumptions,
• increases size when the edge is strong and reduces it when the edge weakens.

Key drawback: full Kelly can be aggressive and is highly sensitive to bad inputs. If your win rate, payoff, or costs are even slightly misestimated, the position size can be too large.

Fractional Kelly (Often The Practical Middle Ground)

Many experienced traders use Kelly as a reference point, then apply a fraction of it:

• ½ Kelly: reduces volatility materially while keeping the adaptive sizing logic
• ¼ Kelly: further dampens drawdowns and makes sizing more resilient to estimation error

Fractional Kelly is often easier to live with because it recognises a reality of trading: your “edge” is not a fixed number — it is an estimate that can change over time.

A Simple Decision Rule

• If you want simplicity and consistency, fixed-percentage risk is usually the more practical baseline.
• If you have robust data and a strategy with measurable, repeatable statistics, Kelly Criterion (especially fractional Kelly) can be a structured way to size dynamically.
• If you are unsure about your inputs, treat full Kelly as a warning sign to reduce size, not a target to hit.

Not sure which approach suits you? Test both on an AvaTrade demo account: run the same strategy with fixed 1% risk, then with ¼ Kelly and ½ Kelly.

Compare equity curve smoothness and maximum drawdown before deciding what’s realistic for your risk tolerance.

Portfolio Risk and Correlation

Kelly is often discussed as if each trade is a standalone bet. In real trading, it rarely is.

Even if every individual trade looks reasonable in isolation, a portfolio can become risky when positions are correlated — meaning they tend to move together, especially during periods of market stress.

Why Correlation Changes the Risk

If you take multiple trades that share the same underlying driver, you may be unintentionally concentrating exposure. For example:

• several FX trades linked to the same currency theme,
• multiple equity positions that behave like the same sector bet,
• risk assets that sell off together during a sudden “risk-off” move.

In these situations, losses can cluster. The result is that your true portfolio risk is higher than your position sizes suggest.

What This Means for Kelly

A Kelly fraction calculated for one opportunity can become misleading when:

• you run several similar trades at the same time, or
• correlations increase unexpectedly (a common feature of volatile markets).

This is one of the strongest arguments for fractional Kelly: it builds a margin of safety for the fact that real-world portfolios are messy, dynamic, and often interconnected.

A Practical Way to Apply This Without Complex Maths

You don’t need portfolio-level equations to act responsibly. A simple risk framework can help:

• Limit the number of “similar” trades you hold simultaneously.
• Cap total risk across correlated positions (e.g., treat several related trades as one theme).
• Reduce size when volatility spikes or when correlations appear to rise.
• Stress-test your portfolio by asking: “What happens if all my positions move against me at the same time?”

The goal is to avoid the scenario where each trade is “properly sized”, yet the combined exposure produces a drawdown you did not plan for.

Common Mistakes When Using the Kelly Criterion

The Kelly Criterion is powerful, but it is also easy to misuse. Most problems do not come from the formula itself — they come from how traders estimate inputs and translate the output into real-world position sizes.

Treating Full Kelly as a “Target”

Full Kelly can produce large position sizes even with reasonable-looking stats. That does not mean it is a sensible size for a live account.

Many traders use fractional Kelly (such as ½ or ¼) specifically to reduce volatility and drawdown risk when the future does not behave like the past.

Feeding the Formula Over-Optimistic Inputs

Small sample sizes, selective backtests, and ignoring trading costs can inflate W and R. When that happens, Kelly does what it is designed to do — it sizes up — but on a false edge.

Forgetting That Correlation Clusters Losses

Applying Kelly “per trade” without considering portfolio exposure can lead to accidental concentration.

In stressed markets, correlations often rise, meaning multiple positions can move against you at the same time.

Ignoring the “No Trade” Signal

If your Kelly fraction is zero or negative, that is not a prompt to adjust the formula — it is a signal that your estimated edge is not positive.

The disciplined response is to reduce risk dramatically or stand aside until the edge is clearer.

Putting Kelly Percentage into a Practical Trading Plan

For most traders, the most effective way to use Kelly is as a framework, not a strict rule.

Step 1: Measure Your Inputs Properly

Build W and R from a meaningful set of trades, including realistic costs and execution. Recalculate periodically to reflect changing conditions.

Step 2: Default To Fractional Kelly

Fractional Kelly helps control volatility and protects you from the reality that estimates are imperfect.

Many professional discussions of Kelly emphasise the practical case for scaling down versus applying full Kelly mechanically.

Step 3: Add Portfolio Guardrails

Even without complex portfolio maths, you can materially reduce risk by:

• capping exposure across related trades,
• limiting the number of similar positions open at once,
• reducing size when volatility spikes or correlations tighten.

Step 4: Stress-Test the “Bad Month”

Before using any sizing method live, ask what happens if:

• your win rate drops temporarily,
• losses arrive in clusters,
• or a tail event slips past your typical stop behaviour.

If the plan does not survive that scenario, the size is too large — regardless of what the formula says.

Build your sizing plan in stages. Use an AvaTrade demo account to test a conservative baseline (fixed % risk), then layer in ¼ Kelly and ½ Kelly. Choose the approach you can execute consistently under pressure.

References For Further Reading

If you want to go deeper into the theory and debates around Kelly, these sources are widely cited:

• J. L. Kelly (1956) — the original paper introducing the framework.
• Edward O. Thorp — practical discussion of Kelly and its application in financial contexts.
• William T. Ziemba — detailed perspective on benefits, riskiness, and discussion of objections (including Samuelson’s).

 Prefer learning by doing? Open an AvaTrade demo account and practise logging trades so you can calculate W and R from real behaviour, not assumptions.

Frequently Asked Questions

• Is The Kelly Criterion “safe” for trading?

  It can be useful, but full Kelly can be aggressive. Many traders use fractional Kelly to reduce drawdowns and estimation risk.

   
• What does a negative Kelly number mean?

  It typically means your estimated edge is not positive. In practice, that is a signal to avoid the trade or improve the strategy.

   
• Is Kelly better than risking 1% per trade?

  Not universally. Fixed-percentage risk is simpler and more robust. Kelly can be helpful if you have reliable statistics and apply conservative fractions.

   
• Can I use Kelly if my trades are correlated?

  You can, but you should be more conservative. Correlation can cause losses to cluster, increasing drawdowns versus what per-trade sizing implies.

   

** Disclaimer – While due research has been undertaken to compile the above content, it remains an informational and educational piece only. None of the content provided constitutes any form of investment advice.