Kelly Criterion - The Mathematics of Position Sizing
By EC Assets Research Team, Risk Strategy · Published · Updated
Kelly Criterion: The Kelly Criterion sizes a position as edge over odds: f = (bp - q) / b, the fraction of capital that maximises long-run geometric growth. A 55 percent win rate at even odds stakes 10 percent. Most practitioners trade half Kelly, because the formula assumes probabilities you rarely know.
What the Kelly Criterion Says
The Kelly criterion answers one question: if an investor faces the same favourable bet many times and reinvests the proceeds, what fraction of capital should go into each bet to make wealth grow fastest in the long run? John Kelly, a researcher at Bell Labs, derived the answer in 1956 in a paper about the rate at which information can be transmitted over a noisy channel; the mathematics turned out to describe betting and investing as well.
The answer is to maximise the expected logarithm of wealth, not expected wealth. Expected wealth is maximised by staking everything on every favourable bet, which guarantees eventual ruin. The logarithm rewards compounding: a 50 percent loss followed by a 50 percent gain leaves capital at 75 percent, and only a growth-rate criterion sees that.
For a bet that pays b times the stake with probability p and loses the stake otherwise, the Kelly fraction is f* = (bp - q) / b. A 55 percent win rate at even odds stakes 10 percent.
How It Works
The expected log growth per bet at stake f is
$$g(f) = p \ln(1 + bf) + q \ln(1 - f)$$
and setting its derivative to zero gives f* = (bp - q) / b. In the continuous approximation, betting a fraction c of the Kelly stake earns c(2 - c) of the maximum growth rate. Measured in multiples of the Kelly stake, which is how estimation errors arrive, the penalty is lopsided: half the stake costs a quarter of the growth, twice the stake costs all of it.
| Stake as share of Kelly | Share of maximum growth | Chance of ever halving capital |
|---|---|---|
| Quarter Kelly | 44% | 0.8% |
| Half Kelly | 75% | 12.5% |
| Full Kelly | 100% | 50% |
| Double Kelly | 0% | 100% |
Continuous-time approximation. The probability of ever falling to half of capital is 0.5 raised to the power 2/c - 1.
For investment returns with mean excess return μ - r and variance σ², the same logic gives f* = (μ - r) / σ², and the growth rate achieved at that fraction is half the square of the Sharpe ratio. Merton reached the same formula in 1969 for an investor with logarithmic utility.
Worked Example
Take the 55 percent coin at even odds and play it 100 times. In the median outcome the investor wins 55 flips and loses 45, so final wealth is (1 + f)^55 × (1 - f)^45:
| Stake per flip | Median wealth after 100 flips | Expected log growth per flip |
|---|---|---|
| 5% (half Kelly) | 1.46× | 0.375% |
| 10% (full Kelly) | 1.65× | 0.501% |
| 20% (double Kelly) | 0.99× | -0.014% |
| 30% | 0.20× | -1.620% |
Every flip has the same positive expected value; the larger stakes lose money only through compounding.
Doubling the stake from 10 to 20 percent doubles the expected profit of each flip and still leaves the median investor slightly poorer after 100 flips. This is volatility drag at work, and the volatility drag calculator shows the same wedge between arithmetic and compounded return.
The same arithmetic applied to equities is sobering. With an excess return of 5 percent and volatility of 18 percent, the Kelly fraction is 0.05 / 0.18² = 1.54: full Kelly would hold 154 percent of capital in equities, half Kelly 77 percent. Kelly is a formula for aggression, which is why it serves better as a ceiling than as a target.
When It Applies (and Limitations)
It needs the true edge, which nobody has. If the coin actually wins 52 percent of the time, the correct stake is 4 percent. An investor who believes in 55 percent stakes 10 percent, two and a half times Kelly, and has negative expected growth of about -0.10 percent per flip. Overestimating the edge is far costlier than underestimating it.
It assumes many repeated, independent bets. For a one-off decision the long-run growth argument does not apply, and with correlated positions the stakes have to be solved jointly, not one at a time.
Its drawdowns are severe. Even with perfect inputs, a full-Kelly bettor has a 50 percent chance of seeing capital halve at some point. Few institutions survive their own governance through a drawdown of that size.
Fat tails shrink the answer. The continuous formula assumes normal returns. Strategies with large occasional losses need a smaller stake than μ / σ² suggests, because the logarithm punishes the rare large loss heavily.
Why It Matters for Institutional Investors
It defines the upper bound on sensible size. Any allocation above Kelly lowers long-run growth while raising risk, so it can only be justified by goals other than growth. Knowing where the bound sits makes the conversation about leverage concrete.
It links sizing to the Sharpe ratio. Because the Kelly growth rate is half the squared Sharpe ratio, a strategy with twice the Sharpe ratio supports four times the growth at its optimal size. Sizing and risk-adjusted return are the same question seen from two sides.
It explains why volatility targeting works. The Kelly fraction can be written as the Sharpe ratio divided by volatility. If the Sharpe ratio stays roughly constant, holding risk constant as volatility changes, as volatility targeting does, keeps the position at a fixed fraction of its Kelly size.
It rewards humility. Because the cost of over-betting is asymmetric, the professional convention of half Kelly or less is not timidity but the rational response to uncertain inputs.
References
- Kelly, J. L. (1956). A New Interpretation of Information Rate. Bell System Technical Journal, 35(4), 917-926.
- Breiman, L. (1961). Optimal Gambling Systems for Favorable Games. Proceedings of the Fourth Berkeley Symposium on Mathematical Statistics and Probability, 1, 65-78.
- Merton, R. C. (1969). Lifetime Portfolio Selection under Uncertainty: The Continuous-Time Case. Review of Economics and Statistics, 51(3), 247-257.
- Thorp, E. O. (2006). The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market. In S. A. Zenios & W. T. Ziemba (Eds.), Handbook of Asset and Liability Management, Vol. 1. Elsevier.
- MacLean, L. C., Thorp, E. O., & Ziemba, W. T. (Eds.). (2011). The Kelly Capital Growth Investment Criterion. World Scientific.
Frequently asked questions
What is the simplest example of the Kelly criterion?
A coin that lands in your favour 55 percent of the time, paying even money. Kelly stakes (1 × 0.55 - 0.45) / 1 = 10 percent of capital on each flip. Over many flips no other fixed fraction grows wealth faster.
Why do practitioners use fractional Kelly?
Because the inputs are estimates. Betting more than Kelly is far more damaging than betting less: half Kelly gives up about a quarter of the growth, while double Kelly gives up all of it. Half Kelly also cuts the probability of ever losing half the bankroll from 50 percent to 12.5 percent.
What happens if the edge is overestimated?
Suppose the true win rate is 52 percent rather than 55. The correct stake is 4 percent, but the 55 percent estimate stakes 10 percent, two and a half times Kelly. At that size the expected log growth is negative, about -0.10 percent per bet, so wealth shrinks over time despite a genuine edge.
How does Kelly apply to investment portfolios?
With continuously distributed returns the Kelly fraction is the expected excess return divided by the variance, the same answer Merton derived for an investor with logarithmic utility. Its growth rate above cash is half the squared Sharpe ratio, which links position sizing directly to risk-adjusted return.
Has Kelly been used by professional investors?
Edward Thorp, who first applied it to blackjack, carried it into securities markets with his convertible and warrant arbitrage at Princeton-Newport Partners from 1969 to 1988, and wrote extensively on its use. Most professional applications use it as an upper bound on size rather than as the size itself.
Stay informed
Market commentary, firm news and research from EC Assets - direct to your inbox.