Home / Position Sizing / Kelly Criterion
Position sizingAdvanced

Kelly Criterion

The Kelly criterion is the bet fraction that maximises the long-run geometric growth rate of capital given a known edge, computed as f* = W − (1 − W) ÷ R, where W is win probability and R is the win-to-loss payoff ratio.

Quick answer: The Kelly criterion is the bet fraction that maximises the long-run geometric growth rate of capital given a known edge, computed as f* = W − (1 − W) ÷ R, where W is win probability and R is the win-to-loss payoff ratio.

In simple words

The Kelly criterion answers a precise question: if you knew your exact edge, what fraction of your capital should you bet to grow it fastest over the long run? The formula rewards a higher win rate and a higher reward-to-risk with a larger fraction. The catch is that it assumes you truly know your edge, which traders never do, and the growth-optimal fraction is also brutally volatile, producing deep drawdowns. In practice full Kelly is treated as an upper bound to stay well below, not a target.

Purpose

Kelly exists to define the mathematically growth-optimal bet size for a known edge, giving a principled ceiling that exposes when a trader is sizing far too large or, occasionally, too timidly.

Visual explanation

Kelly Criterion

Long-run growth rate peaking at the Kelly fraction and turning negative as bets exceed it.

Kelly Fraction & GrowthCapital (log) →Time / trades →Half KellyFull KellyOver-betting

Professional explanation

What Kelly optimises and why it is special

The Kelly criterion maximises the expected logarithm of wealth, which is equivalent to maximising the long-run compound growth rate of capital. This is a stronger and more relevant objective than maximising expected wealth, because compounding is multiplicative: a strategy that occasionally risks too much and suffers a large loss compounds worse even if its average outcome looks high. Kelly is provably growth-optimal in the long run for a known, repeated edge, and betting more than Kelly lowers growth while adding risk, a strictly worse position. This uniqueness is why the criterion is a reference point in both gambling and quantitative finance.

The formula and what each term means

For a simple bet, the Kelly fraction is f* = W − (1 − W) ÷ R, where W is the probability of a win and R is the ratio of the average win to the average loss. If W is 0.5 and R is 2, then f* = 0.5 − 0.5 ÷ 2 = 0.25, meaning bet 25 percent of capital. The formula makes the drivers explicit: a higher win probability or a higher payoff ratio increases the optimal fraction, and when the edge disappears, W − (1 − W) ÷ R falls to zero or below, telling you not to bet at all. A negative Kelly fraction is the maths refusing a negative-expectancy bet.

Why full Kelly is too aggressive for trading

Kelly is growth-optimal only under assumptions traders cannot meet: a known and stable edge, correctly estimated W and R, independent bets, and the ability to bet exact fractions. In reality edges are estimated with error, and Kelly is acutely sensitive to that error, overestimating W or R pushes the recommended fraction far too high. Even with perfectly known parameters, full Kelly produces stomach-churning volatility, with frequent drawdowns of 50 percent or more, because it is optimising growth with no regard for the path. For these reasons practitioners treat full Kelly as a theoretical ceiling and bet a fraction of it.

Estimation error is the fatal practical problem

The inputs W and R are not given; they are estimated from a finite, noisy trade history that may not represent the future. Overestimating the edge, easy to do after a lucky run, inflates the Kelly fraction and can push sizing into the region where growth is actually negative. Because the growth curve falls steeply beyond the Kelly peak, erring on the high side is far more damaging than erring low. This asymmetry, combined with unavoidable estimation error, is the core reason disciplined traders deliberately size below the Kelly figure rather than at it.

Kelly with continuous returns and correlation

The simple win-loss formula is a special case; for continuous returns the Kelly fraction is approximately the expected excess return divided by the variance of returns, which connects it to volatility-based sizing. When multiple positions are held, the single-bet formula breaks down because correlated bets share risk, and the correct multi-asset Kelly fraction depends on the full covariance structure, generally requiring smaller individual bets than the standalone figures suggest. Ignoring correlation and applying single-bet Kelly to each of several aligned positions is a classic way to end up massively over-bet at the portfolio level.

How to use Kelly responsibly

The practical use of Kelly is diagnostic and directional rather than prescriptive. Computed honestly, it tells you the ceiling implied by your estimated edge, and if your intended size is above full Kelly you are certainly over-betting. Most practitioners then apply a fractional Kelly, commonly one-half or less, to cut volatility and buy robustness against estimation error, accepting a modest reduction in theoretical growth for a large reduction in drawdown. Kelly is best seen as a lens that disciplines size, not a dial to set to its maximum.

Formula

f* = W − (1 − W) ÷ R

f* = the Kelly fraction, the share of capital to risk; W = probability of a winning trade (0 to 1); R = payoff ratio, the average win divided by the average loss (reward-to-risk). If f* is zero or negative there is no edge and the maths says do not bet. Example: W = 0.5, R = 2 gives f* = 0.5 − 0.25 = 0.25, i.e. 25 percent, a figure most traders would then halve or quarter.

Full Kelly vs a conservative fixed-fractional approach

AspectFull KellyConservative fixed fractional
ObjectiveMaximise long-run growthCap loss to a small fraction
Inputs neededAccurate W and RA risk percent and a stop
Sensitivity to errorExtreme, punishes overestimated edgeMild
Typical drawdownsVery deep, 50 percent or moreShallow by design
Practical stanceCeiling to stay belowEveryday working rule

Practical example

Illustrative example (Indian market)

A trader estimates from a long history that a Nifty setup wins 50 percent of the time with an average win twice the average loss, so W is 0.5 and R is 2. Kelly gives f* = 0.5 − 0.5 ÷ 2 = 0.25, suggesting risking 25 percent of Rs 5,00,000, or Rs 1,25,000, per trade. That is enormous, a handful of losers would gut the account, and it assumes the 50 percent and the 2 are exactly right, which they are not. Recognising the estimation error and the punishing volatility, the trader applies half-Kelly or less, risking perhaps 6 to 12 percent at most, and in practice caps far lower still at 1 to 2 percent because the edge estimate is uncertain and positions are correlated. Kelly here reveals the ceiling, not the recommendation.

Applying single-bet Kelly to each of several correlated Bank Nifty and Nifty positions at once is a trap: the formula assumes one independent bet, so summing separate Kelly fractions across aligned index trades can imply risking a large multiple of prudent exposure. On a Rs 5,00,000 F&O account the correlation-aware size is far smaller than the standalone numbers suggest.

Advantages

  • Defines the mathematically growth-optimal bet size for a known edge
  • Provides a principled ceiling that exposes gross over-betting
  • Refuses negative-expectancy bets automatically via a non-positive f*
  • Links naturally to volatility sizing through the return-over-variance form
  • Scales the bet up with a stronger edge and down with a weaker one

Limitations

  • Assumes a known, stable edge that traders never actually have
  • Acutely sensitive to overestimated W or R, which pushes size dangerously high
  • Full Kelly produces very deep drawdowns, ignoring the path of returns
  • The simple formula ignores correlation across simultaneous positions
  • Requires whole-lot and margin realities the continuous maths does not model

Common mistakes

  • Betting full Kelly as if the estimated edge were known exactly
  • Overestimating win rate or payoff after a lucky run and over-sizing
  • Applying the single-bet formula to several correlated positions at once
  • Treating a negative Kelly output as a small bet rather than no bet
  • Confusing the concept page with the calculator and skipping honest inputs
  • Ignoring that full Kelly can imply drawdowns most traders cannot tolerate

Professional usage

Quantitative desks use Kelly as an analytical benchmark rather than a live throttle. They compute it from carefully estimated, out-of-sample edge parameters, use the covariance-aware multi-asset form when several positions interact, and then deliberately run at a fraction of Kelly, often a quarter to a half, to survive estimation error and cap drawdown. The prevailing view is that being at or above full Kelly is a red flag, while being well below it trades a little growth for a lot of robustness, which is usually the right bargain.

Key takeaways

  • Kelly maximises long-run growth for a known edge: f* = W − (1 − W) ÷ R
  • It is extremely sensitive to estimation error and over-betting is severely punished
  • Full Kelly implies drawdowns most traders cannot stomach
  • Use it as a ceiling and run a fraction of it, not the full figure

Frequently asked questions

What is the Kelly criterion?
The Kelly criterion is the bet fraction that maximises the long-run compound growth rate of capital for a known, repeated edge. It is computed as f* = W minus (1 minus W) divided by R, where W is the win probability and R is the ratio of average win to average loss.
What is the Kelly formula?
For a simple win-or-lose bet, f* = W − (1 − W) ÷ R. W is the probability of winning and R is the payoff ratio, average win over average loss. If W is 0.5 and R is 2, f* is 0.25, meaning the growth-optimal bet is 25 percent of capital.
Why should I not bet full Kelly?
Because full Kelly assumes you know your edge exactly, which you do not, and it produces very deep drawdowns since it optimises growth without regard to the path. Overestimating your edge pushes the fraction dangerously high, so most traders run a fraction of Kelly instead.
What does a negative Kelly fraction mean?
A negative or zero f* means the bet has no positive edge, so the mathematics says do not bet at all. It is not an instruction to bet small; it is the formula refusing a neutral or negative-expectancy trade.
How sensitive is Kelly to my inputs?
Extremely. The recommended fraction moves sharply with small changes in the estimated win rate or payoff ratio, and because the growth curve falls steeply past the Kelly peak, overestimating your edge is far more damaging than underestimating it. That asymmetry is why traders size below Kelly.
Does Kelly account for multiple positions?
The simple formula does not; it assumes a single independent bet. With several positions the correct fraction depends on the full covariance structure and is generally smaller than the standalone figures, so applying single-bet Kelly to each of several correlated trades badly over-bets the portfolio.
How does Kelly relate to volatility sizing?
For continuous returns, the Kelly fraction is approximately the expected excess return divided by the variance of returns. This connects Kelly directly to volatility-based sizing, where position size falls as variance rises, and it shows the two ideas are facets of the same growth-optimal logic.
Is Kelly useful if I cannot know my edge?
Yes, diagnostically. Even a rough, honest estimate reveals a ceiling: if your intended size exceeds full Kelly, you are certainly over-betting. Kelly is best used to detect gross over-sizing rather than to set an exact live bet.
What drawdowns does full Kelly produce?
Full Kelly commonly produces drawdowns of 50 percent or more, because it maximises growth with no penalty for the path taken. Such drawdowns are psychologically and practically intolerable for most traders, which is another reason fractional Kelly is preferred.
How is Kelly different from the 1 percent rule?
The 1 percent rule caps loss at a small fixed fraction for survival, while Kelly computes a growth-optimal fraction from an estimated edge, which is usually far larger. Kelly answers how much maximises growth; the 1 percent rule answers how little keeps you safe. Most traders sit far closer to the latter.
Can I use Kelly for option selling?
With great caution. Option selling often has a high win rate and a poor payoff ratio, and a fat-tailed loss distribution that violates Kelly's simple assumptions. The formula can suggest a large fraction while the true tail risk makes even a small size dangerous, so naive Kelly is especially misleading here.
Why does Kelly maximise the log of wealth?
Because compounding is multiplicative, and maximising expected log wealth is equivalent to maximising the long-run geometric growth rate. This penalises large losses appropriately, since a severe drawdown hurts compounding more than an equal gain helps, which simple expected-wealth maximisation ignores.
What fraction of Kelly do people actually use?
Commonly one-half or less, and often far less once estimation error and correlation are considered. Half-Kelly retains most of the theoretical growth while roughly halving volatility, and many practitioners cap real risk near 1 to 2 percent regardless of what Kelly suggests.
Is the Kelly concept the same as the Kelly calculator tool?
They are related but distinct. This page explains the concept, its assumptions and dangers, while the calculator lets you plug in W and R to see the fraction. The tool is only as trustworthy as the honesty of the edge estimates you feed it.

Voice search & related questions

Natural-language questions people ask about Kelly Criterion.

What is the Kelly criterion?
It is a formula that says how big to bet to grow your money fastest, if you truly knew your edge. In practice you bet much less because you never know it exactly.
What is the Kelly formula?
It is f-star equals win probability minus one minus win probability, divided by your reward-to-risk ratio. A bigger edge means a bigger suggested bet.
Why is full Kelly dangerous?
Because it assumes you know your edge perfectly and it swings your account very hard, with drawdowns of half or more. Most people bet a fraction of it.
What does a negative Kelly number mean?
It means the bet has no real edge, so the maths is telling you not to take the trade at all, not to bet a tiny amount.
How much Kelly should I actually bet?
Usually half or less, and often far less once you allow for uncertainty and correlated trades. Many traders cap real risk near one or two percent anyway.
Is Kelly better than the one percent rule?
They answer different questions. Kelly chases fastest growth; the one percent rule protects you. Most traders lean on the safer rule and use Kelly as a ceiling.

Sources & references

    Last reviewed 12 July 2026. Educational content only — not investment advice. Markets and rules change; verify current conventions with SEBI, NSE/BSE and your broker.

    ← Previous
    Percentage Risk Model
    Educational content only — not investment advice. Examples use illustrative numbers and simplified models. Risk-management techniques reduce but never remove risk, and trading derivatives involves substantial risk of loss. See our Risk Disclosure and SEBI Disclaimer.