Quant Analytics: from scratch to pro · Portfolio, sizing and risk metrics · Advanced
Position sizing: fixed fractional, volatility targeting and Kelly
How much to put at risk matters as much as what to buy. Three sizing methods, fixed fractional risk, volatility targeting and the Kelly criterion, each answer a different question, and the Kelly growth curve shows why betting too much is worse than betting too little.
In this lesson
- Size a position from an account risk fraction and a stop distance
- Scale exposure to a volatility target
- Compute the Kelly fraction and the growth rate at any bet fraction
- Explain why half Kelly is common and what estimation error does to full Kelly
Read first: Position sizing basics, Risk-adjusted return: the Sharpe and Sortino ratios
Two traders with the same signal and different sizes end up with very different results. Position sizing decides how much of an account rides on each idea. Three methods are in common use. They answer different questions.
Fixed fractional risk
Choose the share of the account you are willing to lose if the trade fails, then back out the size from the stop distance:
size = (account x risk fraction) / stop distance per unit
Take an account of 50,000, a risk fraction of 1% and a stop 2 ATR from entry, where ATR (average true range, a daily range measure) is 2.5. The stop distance is 5 and the risk amount is 500. The position is 500 / 5, which is 100 shares. A wider stop means a smaller position; the loss at the stop stays at 1%. The method controls the loss per trade. It does not say what fraction is right. See position sizing basics and position size.
Volatility targeting
Instead of fixing the loss, fix the portfolio's volatility: exposure = target volatility / asset volatility. With a 10% target and an asset at 20% volatility, exposure is 10% / 20%, which is 50%. If volatility doubles to 40%, exposure falls to 25%. Risk stays steady as markets calm and stir. The cost is trading as volatility moves, and a lag: the estimate is always behind (see the volatility models lesson on EWMA and GARCH). See volatility targeting.
The Kelly criterion
J. L. Kelly (1956) asked what fraction of capital to stake on a repeated favourable bet to maximise long-run growth. For a bet that wins b times the stake with probability p and loses the stake with probability q = 1 - p:
f* = p - q / b
The long-run growth rate per bet at fraction f is g(f) = p ln(1 + b f) + q ln(1 - f). It is the expected log growth, because wealth compounds. See Kelly criterion.
Worked example
A bet pays even money (b = 1) and wins with probability 0.55. Then f* = 0.55 - 0.45 / 1, which is 10%.
| Fraction staked | Growth per bet |
|---|---|
| 5% (half Kelly) | 0.38% |
| 10% (full Kelly) | 0.50% |
| 20% (double Kelly) | -0.014% |
Half Kelly captures 75% of the growth of full Kelly with half the stake, and much smaller swings. Double Kelly earns nothing in the long run, despite the edge. The curve is asymmetric in what matters: staking too little costs growth gradually, staking too much can erase it. That is the lesson of the chart.
Estimation error
The edge is never known, only estimated. Suppose the true win probability is 0.52, but you estimate 0.55 and stake 10%. The true Kelly fraction is 0.52 - 0.48 / 1, which is 4%. Staking 10% is 2.5 times too much, and the growth rate per bet becomes -0.10%. A modest overestimate of the edge turns the best strategy into a losing one. That is why practitioners use a fraction of Kelly (half, or less) and treat the estimate as an upper bound.
For a continuous setting with excess return mu and variance sigma^2, the equivalent is f* = mu / sigma^2. With mu = 5% and sigma = 16% it is 0.05 / 0.0256, which is 1.95 times capital, more leverage than most portfolios carry. See leverage and margin.
Common mistakes
- Betting full Kelly on an estimated edge. Estimation error pushes the stake past the optimum.
- Applying the formula to outcomes that are not repeated independent bets.
- Ignoring that returns are fat-tailed. Kelly assumes the distribution is known.
- Treating the stop loss as a guarantee. Gaps fill beyond it (see gaps and weekend risk).
- Sizing each trade alone and ignoring that open positions are correlated.
- Mixing up the risk fraction per trade with the Kelly fraction. They answer different questions.
In practice
Risk per trade is often set as a small fixed fraction of the account, such as 1% or 2%, with exposure adjusted for volatility and for how many positions move together. Kelly is best used as a ceiling: a way to see how much risk would be too much, not a target to reach. Trading on FinUties starts on a demo account, where ideas like this can be practised without real money.
Long-run growth per bet against the fraction staked
- p = 0.55, even money
- p = 0.52, even money
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| Kelly fraction, p = 0.55, even money | 0.55 - 0.45 / 1 | 10% |
| growth per bet at full Kelly | 0.55 * ln(1.10) + 0.45 * ln(0.90) | 0.50% |
| growth per bet at half Kelly | 0.55 * ln(1.05) + 0.45 * ln(0.95) | 0.38% |
| half Kelly growth as a share of full | g_half / g_full | 75% |
| growth per bet at double Kelly | 0.55 * ln(1.20) + 0.45 * ln(0.80) | 0.014% |
| true Kelly fraction when p = 0.52 | 0.52 - 0.48 / 1 | 4% |
| growth betting 10% when p = 0.52 | 0.52 * ln(1.10) + 0.48 * ln(0.90) | 0.10% |
| shares for 1% risk of 50,000 with a stop 5 away | 0.01 * 50000 / 5 | 100 |
| exposure for a 10% target at 20% volatility | 0.10 / 0.20 | 50% |
| exposure when volatility rises to 40% | 0.10 / 0.40 | 25% |
| continuous Kelly leverage, excess return 5%, volatility 16% | 0.05 / 0.16**2 | 1.95 |
Check yourself
A bet pays even money and wins 60% of the time. What is the Kelly fraction?
f = p - q / b = 0.60 - 0.40 / 1 = 0.20, so 20% of capital per bet.
An account of 20,000 risks 1% per trade and the stop is 4 below entry. How many shares?
Risk amount 200, divided by 4 per share, is 50 shares.
An asset has 25% annual volatility and the target is 10%. What is the exposure?
10% / 25% = 0.40, so 40% of capital, with the rest in cash.
Why does betting twice the Kelly fraction give roughly zero growth?
Near the optimum the growth curve is a downward parabola. At twice the optimum it has fallen back to about zero, and beyond that it is negative: more risk, lower growth.
The true win probability is 52% but you estimated 55% and bet the Kelly fraction for 55% (10%). What happens?
The true Kelly fraction is 4%. Betting 10% is 2.5 times too much and the long-run growth rate is negative.
Key terms
Kelly criterionVolatility targetingPosition sizeRisk of ruin
Sources
- A New Interpretation of Information Rate. Bell System Technical Journal 35(4), 917-926 (J. L. Kelly Jr.), 1956. Accessed 9 Nov 2026
- The Kelly Capital Growth Investment Criterion: Theory and Practice. World Scientific (Leonard C. MacLean, Edward O. Thorp, William T. Ziemba, eds.), 2011. Accessed 9 Nov 2026