Kelly Criterion: Optimal Position Sizing for Traders
The Kelly Criterion is one of the most important concepts in position sizing and money management. Whether you’re a systematic trader, a sports bettor, or an investor sizing positions in a portfolio, understanding Kelly can help you maximize long-term wealth growth while avoiding the extremes of betting too conservatively or too aggressively. This guide covers the Kelly formula, how to calculate optimal position sizes, when to use fractional Kelly, and the critical limitations every practitioner should understand.
What Is the Kelly Criterion?
The Kelly Criterion is a mathematical formula that determines the optimal fraction of capital to risk on each bet or trade to maximize long-term geometric growth. Developed by John Kelly at Bell Labs in 1956 and later popularized by mathematician Ed Thorp (who used it to beat casinos at blackjack and later manage a successful hedge fund), Kelly remains a cornerstone of quantitative money management.
The Kelly Criterion maximizes expected log-wealth (geometric mean return), not expected dollar wealth. This distinction matters: maximizing expected wealth can lead to ruin through overbetting, while Kelly finds the sweet spot that grows wealth fastest over time without risking everything.
The intuition is straightforward: bet too little and your wealth grows slowly; bet too much and a string of losses will devastate your capital. Kelly identifies the fraction that balances these extremes — the point where your long-run compound growth rate is highest.
Kelly is not about maximizing expected profit on any single bet. Instead, it’s about maximizing the expected growth rate across many repeated bets, which is equivalent to maximizing the geometric mean of outcomes.
The Kelly Criterion Formula
The Kelly Criterion has two primary forms: one for discrete binary outcomes (common in gambling) and one for continuous returns (common in investing).
In the discrete formula:
- p — probability of winning
- q — probability of losing (q = 1 – p)
- b — net odds received on the bet (profit per $1 wagered if you win, not including your stake back)
The variable b represents net odds, meaning profit per unit staked. If you bet $100 and win $150 (getting back $250 total), then b = 1.5. This is a common source of confusion — always use net profit, not gross payout.
In the continuous formula:
- μ — expected return of the risky asset (annualized)
- r — risk-free rate
- σ — standard deviation (volatility) of the risky asset
- f* — optimal fraction to allocate to the risky asset
The continuous version is derived from a geometric Brownian motion model with continuous rebalancing and no transaction costs. It assumes returns follow a diffusion process — a stylized but useful approximation for many financial applications.
Interpreting Kelly Criterion Values
The Kelly fraction can take on any value, and understanding what different ranges mean is essential for practical application:
| Kelly Fraction | Interpretation | Action |
|---|---|---|
| f* < 0 | Negative expected value | Don’t bet (or short if allowed) |
| f* = 0 | Zero edge | No advantage to betting |
| 0 < f* < 1 | Positive edge, no leverage needed | Bet this fraction of your capital |
| f* = 1 | Optimal to bet everything | Rare; requires extreme edge or low vol |
| f* > 1 | Model recommends leverage | Use with extreme caution (see below) |
When f* exceeds 100%, the Kelly model is recommending leverage — borrowing to increase exposure. This is an unconstrained mathematical output, not a practical recommendation. Real-world constraints (margin limits, borrowing costs, psychological tolerance) usually make full Kelly leverage inadvisable.
How to Calculate Kelly Criterion Position Size
Let’s work through two examples: one discrete (trading) and one continuous (investing).
A trader specializing in NVIDIA (NVDA) earnings announcements has tracked their strategy over 20 quarters:
- Win rate: 60% (12 wins out of 20 trades)
- Average win: $1,500 per contract
- Average loss: $1,000 per contract
First, calculate the net odds ratio:
b = Average Win / Average Loss = $1,500 / $1,000 = 1.5
Now apply the Kelly formula:
f* = (b × p – q) / b = (1.5 × 0.60 – 0.40) / 1.5 = (0.90 – 0.40) / 1.5 = 0.50 / 1.5 = 0.333
Result: Kelly suggests risking 33.3% of capital on each NVDA earnings trade. Most traders would use half-Kelly (16.7%) or quarter-Kelly (8.3%) in practice, given the uncertainty in estimating future win rates from historical data.
An investor is deciding how much to allocate to the SPDR S&P 500 ETF (SPY) versus Treasury bills, using historical data:
- SPY expected return (μ): 10% annually (long-term historical average)
- 3-month T-bill rate (r): 4.5%
- SPY volatility (σ): 16% (historical standard deviation)
Apply the continuous Kelly formula:
f* = (μ – r) / σ2 = (0.10 – 0.045) / (0.16)2 = 0.055 / 0.0256 = 2.15
Result: Kelly suggests 215% allocation to SPY — implying 115% leverage. This illustrates why the continuous formula often produces theoretical results that exceed practical constraints. A half-Kelly allocation of 107% would require modest leverage, while quarter-Kelly (54%) would be achievable in a standard brokerage account without margin.
The second example demonstrates why full Kelly is rarely used in practice: the model’s assumptions (perfect parameter estimates, continuous rebalancing, no transaction costs) don’t hold in the real world.
Fractional Kelly: Half-Kelly and Quarter-Kelly
Because full Kelly requires knowing true probabilities and payoffs — which are always estimated with error — most practitioners use a fraction of the Kelly bet. This is called fractional Kelly.
| Strategy | Position Size | Expected Growth | Volatility Reduction |
|---|---|---|---|
| Full Kelly | f* | 100% | 0% |
| Half Kelly | f* / 2 | 75% | 50% |
| Quarter Kelly | f* / 4 | ~44% | 75% |
Half-Kelly cuts your volatility in half while only reducing expected growth by 25%. This is an exceptionally favorable trade-off, especially when your probability and payoff estimates contain uncertainty.
The mathematics of fractional Kelly reveal an important insight: the growth-rate curve is relatively flat near the optimum. Betting slightly less than Kelly costs little in expected growth but dramatically reduces variance and drawdown risk.
Beyond half Kelly, going to the “crazy zone” — betting more than Kelly — is always suboptimal: you get lower expected growth and higher volatility. At exactly twice Kelly, your excess growth rate above the risk-free rate drops to zero. Beyond twice Kelly, expected growth actually becomes negative.
Kelly Criterion vs Fixed Fractional Sizing
The most common alternative to Kelly is fixed fractional sizing, where you risk a constant percentage of capital on every trade regardless of the expected edge.
Kelly Criterion
- Edge-responsive: Larger bets when edge is bigger
- Maximizes long-run geometric growth rate
- Requires accurate probability and payoff estimates
- Can produce large position sizes (even leverage)
- Best for: Known edges, systematic strategies
Fixed Fractional
- Rule-based: Same percentage every trade (e.g., 2%)
- Does not maximize growth but limits drawdowns
- No probability estimation required
- Simple to implement and psychologically easier
- Best for: Uncertain edges, discretionary trading
The main advantage of fixed fractional sizing is that it avoids the parameter-estimation burden that makes full Kelly fragile. If you don’t know your true edge, a conservative fixed fraction may outperform a Kelly calculation based on inaccurate inputs.
Kelly Criterion in Trading and Investing
Kelly has been applied successfully across multiple domains:
- Professional gambling: Ed Thorp famously used Kelly to beat blackjack casinos and later founded Princeton Newport Partners, one of the first quantitative hedge funds
- Systematic trading: Algorithmic traders use Kelly to size positions based on backtested win rates and payoff ratios
- Sports betting: Sharp bettors apply Kelly when they have a quantifiable edge over the bookmaker’s odds
- Portfolio allocation: The continuous Kelly formula relates to mean-variance optimization and can inform stock/bond allocation decisions
There’s an elegant connection between Kelly and the Sharpe ratio. For a single risky asset, the Kelly fraction can be written as:
f* = (μ – r) / σ2 = Sharpe Ratio / σ
This shows that Kelly sizing increases with the Sharpe ratio and decreases with volatility. The Kelly-optimal excess growth rate above the risk-free rate equals Sharpe2/2. The total expected log-growth rate is r + Sharpe2/2, which provides a useful benchmark for comparing strategies.
Limitations of the Kelly Criterion
While mathematically elegant, Kelly has significant practical limitations that every user should understand:
Kelly assumes you know the true probabilities and payoffs. In practice, you’re always estimating these parameters. Overestimate your edge and Kelly will tell you to overbet — potentially leading to ruin.
1. Estimation Error — Garbage in, garbage out. Small errors in probability estimates can produce large errors in position size. This is why fractional Kelly is essential in practice.
2. Non-Independent Returns — Standard Kelly assumes each bet is independent. In trading, positions are often correlated, and losing streaks can cluster. Kelly doesn’t account for this without extensions to multi-asset portfolio Kelly.
3. Drawdown Risk — Full Kelly can produce drawdowns of 50% or more. While the math says you’ll recover, the psychological and practical reality of such drawdowns is devastating for most investors.
4. Two-Outcome Simplification — The discrete Kelly formula assumes binary win/lose outcomes. Real trades have distributions of outcomes, gap risk, and slippage that the simple formula ignores.
5. No Transaction Costs — The continuous Kelly model assumes frictionless rebalancing. Real trading involves commissions, spreads, and market impact that erode returns.
6. Leverage Constraints — When Kelly recommends f* > 100%, you may face real-world limits: margin requirements, borrowing costs, and mandate restrictions that prevent full implementation.
Common Kelly Criterion Mistakes
Understanding these pitfalls will help you apply Kelly more effectively:
Mistake 1: Using Full Kelly with Estimated Parameters
Overconfidence in your probability estimates leads to overbetting. Always use fractional Kelly (half or quarter) unless you have extreme confidence in your edge estimate — which is rare.
Mistake 2: Treating f* > 100% as a Mandate to Leverage
When the Kelly formula outputs a fraction greater than 1, it’s an unconstrained mathematical result. Real constraints (financing, mandates, psychology) should limit actual leverage. Treat high Kelly fractions as a sign of an attractive opportunity, not an instruction to max out margin.
Mistake 3: Ignoring Correlation Between Positions
If you apply Kelly independently to multiple correlated positions, you’ll systematically overbet. Portfolio Kelly requires considering the covariance matrix of returns.
Mistake 4: Confusing Kelly with Sharpe Ratio
Kelly tells you how much to bet going forward; Sharpe ratio measures risk-adjusted return and is often calculated ex post (though it can also be used ex ante in optimization). They’re mathematically related but serve different purposes in portfolio management.
Mistake 5: Applying Binary Kelly to Multi-Outcome Trades
The discrete formula assumes win or lose. If your trades have stop-losses, take-profits, and partial fills, you need to either simulate or use a more sophisticated model.
Mistake 6: Using Kelly When Edge Is Zero or Negative
If f* ≤ 0, don’t bet. Some traders force trades even when their system shows no edge — Kelly explicitly says to stay out.
Frequently Asked Questions
Disclaimer
This article is for educational and informational purposes only and does not constitute investment or trading advice. The Kelly Criterion is a theoretical framework with significant practical limitations. Position sizing should consider your individual risk tolerance, investment constraints, and the uncertainty in your probability estimates. Always conduct your own research and consult a qualified financial advisor before making investment decisions.
