Kelly Criterion

Quick Definition

The Kelly Criterion answers a question every investor faces: how much should I put into this position? Developed by John L. Kelly Jr. at Bell Labs in 1956, the formula calculates the fraction of your capital that maximizes long-term compound growth when you have a measurable edge. Bet too small and you leave gains on the table. Bet too large and a string of losses can wipe you out. Kelly finds the exact fraction that grows your wealth fastest over many repeated bets.

What It Means

Position sizing determines whether a strategy compounds wealth or destroys it. A brilliant investment thesis with a genuine edge can still produce ruin if you commit too much capital. A mediocre strategy with disciplined sizing can outperform a superior strategy with reckless allocation. The Kelly Criterion gives you a mathematical framework for sizing every position based on your confidence in the outcome and the payoff structure.

The formula comes in two forms. The discrete version handles binary outcomes (win or lose a fixed amount):

Kelly Fraction = (b x p - q) / b

Where:

• b = the ratio of the amount you win to the amount you wager (net odds)
• p = probability of winning
• q = probability of losing (1 minus p)

The continuous version handles investments with normally distributed returns:

Kelly Fraction = Mean Return / Variance of Returns

Or equivalently: f = mu / sigma squared, where mu is the expected return and sigma squared is the variance.

The math is straightforward. The difficulty lies in estimating the inputs. You never know your true win rate or your true payoff ratio with certainty. You have a sample of past results, possibly biased, almost certainly drawn from a different regime than the one you will face going forward. This estimation problem is why almost nobody runs full Kelly in practice.

How It Works

Consider a simple example. You find a stock that you believe has a 60% chance of rising 50% and a 40% chance of falling 30%. Should you buy it? Yes, but how much?

Using the discrete Kelly formula:

• b = 50/30 = 1.667 (you win 50 for every 30 you risk)
• p = 0.60
• q = 0.40

Kelly = (1.667 x 0.60 - 0.40) / 1.667 = (1.00 - 0.40) / 1.667 = 0.36

The formula says to put 36% of your portfolio into this trade. That is the bet size that maximizes long-term compound growth.

Now consider what happens at different bet sizes. If you bet 10% of your capital, you grow steadily but slowly. If you bet 36% (full Kelly), you grow at the maximum possible rate. If you bet 50%, you still grow, but slower than at 36%, and with much larger drawdowns. If you bet 60%, your expected geometric growth turns negative. You have a positive expected return on each trade, but your wealth shrinks over time because the losses compound too aggressively.

This is the critical insight: above the Kelly fraction, your expected geometric growth rate goes negative even though each individual bet has positive expected value. The math punishes overbetting severely.

Fractional Kelly

Full Kelly produces drawdowns of 50% to 80% that are mathematically expected, not rare. The ride is brutal. Most practitioners use fractional Kelly, typically half Kelly or quarter Kelly, to reduce volatility at the cost of slightly slower growth.

Half Kelly gives up 25% of your growth rate but cuts your drawdowns roughly in half. For most investors, that trade is worth making. Quarter Kelly gives up 56% of your growth rate but produces a much smoother equity curve that you can actually stick with through difficult periods.

Why Full Kelly Is Rarely Practical

Full Kelly is optimal only under a set of assumptions that never hold in real markets. The formula assumes you know the true probability distribution of outcomes. It assumes your edge is stable over time. It assumes no transaction costs, continuous rebalancing, frictionless margin, and no career risk, redemption pressure, liquidity constraints, tax considerations, or governance limits. Every one of those assumptions is violated in real investing.

A 2026 analysis from Atlas Peak Research concluded that the Kelly Criterion should be treated as a theoretical upper bound on rational position size, not as a practical target. The recommended institutional process is to shrink your expected return estimate first (accounting for parameter uncertainty), compute raw Kelly as a reference point, then impose explicit drawdown constraints, liquidity limits, and governance caps. The binding position size should be the minimum of the adjusted Kelly and your institutional risk limits.

Research from Strasmore tested the Kelly formula against ten years of daily stock returns (2016 through 2025) across six familiar names. The inputs looked tiny: win rates clustered near a coin flip, and average up days and down days were close in size. SPY closed higher on 55.3% of sessions, gaining 0.71% on an average up day against 0.76% on an average down day. But those small, nearly balanced inputs sit in the denominators of the formula, and small denominators produce large answers. Full Kelly prescribed position sizes that were unlivable in practice.

Real-World Examples

Sports Betting

The Kelly Criterion originated in gambling and remains widely used there. A bettor analyzing an NFL game estimates a team has a 55% chance of covering the spread at 2.0 decimal odds (bet 100 to win 100):

• b = 1.0 (even money)
• p = 0.55
• q = 0.45

Kelly = (1.0 x 0.55 - 0.45) / 1.0 = 0.10

Full Kelly says to bet 10% of your bankroll. A half-Kelly bettor would wager 5%. Professional sports bettors almost universally use fractional Kelly, typically quarter Kelly or lower, because their edge estimates are noisy and variance is extreme.

Stock Investing

An investor identifies a stock trading at $30 that they believe is worth $50 based on fundamental analysis. They estimate a 60% chance the stock reaches $50 within two years and a 40% chance it falls to $20:

• b = 20/10 = 2.0 (potential gain of $20 vs potential loss of $10)
• p = 0.60
• q = 0.40

Kelly = (2.0 x 0.60 - 0.40) / 2.0 = (1.20 - 0.40) / 2.0 = 0.40

Full Kelly recommends 40% of portfolio capital. Half Kelly would suggest 20%. Most professional investors would cap this position at 5% to 10% of portfolio capital, reflecting the enormous uncertainty in the probability estimate and the desire to maintain diversification.

Venture Capital

Venture capital returns follow a power law distribution, which makes Kelly sizing particularly challenging. According to iCapital research published in May 2026, just 3% of venture capital deals account for nearly half of the industry's total returns. Correlation Ventures studied over 27,000 financings from 2009 to 2018 and found that 65% of deals returned less than the capital invested, while only 0.4% returned more than 50x. The Kelly formula struggles with these distributions because the payoff ratios are extreme and the probability of the outlier outcomes is nearly impossible to estimate accurately.

Key Points to Remember

• The Kelly Criterion calculates the position size that maximizes long-term compound growth when you have a measurable edge
• Full Kelly is a theoretical ceiling, not a practical target. Drawdowns of 50% to 80% are mathematically expected under full Kelly
• Most practitioners use half Kelly or quarter Kelly to reduce volatility while retaining most of the growth benefit
• The formula is only as good as your input estimates. Overestimating your edge leads to overbetting, which can turn a positive-expectancy strategy into a wealth destroyer
• Above the Kelly fraction, expected geometric growth turns negative even with positive expected value per trade
• Kelly assumes you know the true probability distribution, which you never do in real markets

Common Mistakes to Avoid

• Running full Kelly with estimated inputs: Your edge estimates are always noisy. If you plug overestimated win rates into the Kelly formula, you will overbet. The solution is fractional Kelly combined with conservative input estimates. If your model says 40%, bet 10% to 15%.
• Ignoring estimation error: The Kelly formula assumes known probabilities. In reality, you have a sample of past results that may not represent future outcomes. Bayesian and shrinkage approaches that reduce the expected return input before sizing are more appropriate for institutional investing.
• Confusing arithmetic return with geometric return: A strategy can have positive expected arithmetic return and negative expected geometric growth if position sizes are too large. This is the trap that destroys leveraged traders. The Kelly fraction is the boundary where geometric growth peaks.
• Applying Kelly to non-repeated bets: Kelly optimizes growth over many repeated bets with similar characteristics. A single once-in-a-lifetime investment opportunity does not fit the framework. Kelly is for serial decision-making, not one-off wagers.
• Neglecting behavioral finance realities: Even if you calculate the mathematically optimal position size, you may not be able to hold through the drawdowns. A 50% drawdown under full Kelly causes most investors to abandon the strategy. Fractional Kelly produces drawdowns you can psychologically tolerate.

Related Concepts

The Kelly Criterion sits within the broader framework of risk and return management. It connects directly to alpha, which measures the edge you are trying to size, and beta, which measures your exposure to market movements. Investors who apply Kelly thinking are usually engaged in fundamental analysis to estimate the probability and magnitude of outcomes. The discipline of position sizing also relates to diversification, since Kelly naturally limits concentration when your edge is uncertain. For investors interested in the psychology of holding through drawdowns, behavioral finance explains why most people cannot tolerate full Kelly volatility. For a deeper dive into practical position sizing, read our guide on common investing mistakes beginners make and our explanation of dollar-cost averaging as an alternative to concentrated bets. You can also use our investment return calculator to model how different position sizes affect your portfolio over time. For academic background, the original Kelly paper is available through Bell System Technical Journal archives.

Frequently Asked Questions

Q: Should I use the Kelly Criterion for my retirement portfolio? A: Probably not in its direct form. Retirement investing involves uncertain probabilities over long time horizons, and full Kelly sizing would create intolerable volatility. The useful takeaway from Kelly for retirement investors is the principle: size positions based on your confidence in the edge, and never bet so much that a string of losses becomes catastrophic. Most retirement savers are better served by broad diversification and dollar-cost averaging.

Q: What is the difference between Kelly and fixed-fractional sizing? A: Fixed-fractional sizing means risking the same percentage of your capital on every trade, regardless of your edge. Kelly adjusts the size based on the strength of your edge and the payoff ratio. Fixed-fractional is simpler and more resilient to estimation error. Kelly is theoretically optimal but requires accurate probability estimates that most investors cannot produce.

Q: Why do professional gamblers use Kelly but most investors do not? A: Gamblers operate in environments where probabilities can be estimated with reasonable accuracy (card counting, sports modeling) and outcomes are discrete and frequent. Investors face continuous distributions, regime changes, and noisy edge estimates. The conditions for Kelly to work well are better approximated in gambling than in investing.

Q: What happens if I bet above the Kelly fraction? A: Your expected geometric growth rate turns negative. You may still have positive expected arithmetic return on each individual bet, but the compounding effect of large losses means your wealth shrinks over time. This is why overbetting is more dangerous than underbetting. Underbetting costs you growth. Overbetting costs you everything.