The Kelly Criterion: Optimal Position Sizing for Real Traders
Most traders agonize over what to buy. Professionals know the harder question is how much to buy. Bet too much, and one bad trade ends your career. Bet too little, and you give up serious compounding. The Kelly Criterion answers this question mathematically — and once you understand it, position sizing stops being intuition and becomes arithmetic.
Developed by physicist John L. Kelly Jr. at Bell Labs in 1956 to maximize information-theoretic returns, the formula was famously used by Edward Thorp to beat blackjack and then by his hedge fund Princeton-Newport Partners (which never had a losing year over 19 years). Warren Buffett, despite his folksy aphorisms, sizes his concentrated positions in a way mathematically equivalent to fractional Kelly. This isn't academic curiosity — it's how the best in the world actually allocate capital.
What Is the Kelly Criterion?
Kelly answers a deceptively simple question: given a bet with known probability of winning and known payoff, what fraction of your bankroll should you wager to maximize long-term compounded growth?
= (b × p − q) / b
Where:
- f* = optimal fraction of bankroll to bet
- p = probability of winning
- q = 1 − p = probability of losing
- b = payoff ratio (win size / loss size, "odds")
A Concrete Example
Suppose you've identified a trade setup with a 60% historical win rate. When you win, you make $300. When you lose, you lose $100. The payoff ratio b = 3.
Kelly says: f* = (3 × 0.6 − 0.4) / 3 = 1.4 / 3 = 0.467
You should bet 46.7% of your bankroll on this trade for maximum geometric growth.
That's enormous. And there's a reason almost no one in finance actually uses full Kelly...
Why Full Kelly Will Ruin You
Full Kelly is mathematically optimal only when you know the true win probability and payoff with certainty. In real trading, you don't. Your estimate of "60% win rate" is based on a small sample. The true number might be 50% or 65%. If the true probability is even slightly lower than you estimate, full Kelly can produce massive drawdowns.
Edward Thorp computed that if you bet 1.5× your Kelly amount, your long-term growth rate becomes zero. Bet 2× Kelly, and you guarantee bankruptcy over time, despite having a positive-expectation edge.
The Volatility Problem
Full Kelly produces massive volatility. In the example above (46.7% per trade), a series of three losses in a row would lose 89% of your capital:
(1 - 0.467)³ = 0.151 — only 15% of starting capital left.
Psychologically intolerable. Even if mathematically optimal, you'll abandon the strategy before the long-term growth materializes.
Half-Kelly: What Pros Actually Use
The standard professional approach is Half-Kelly (or Quarter-Kelly for highly uncertain setups). The math:
- Half-Kelly captures ~75% of the optimal growth rate
- ...but with only ~50% of the volatility
- Massive risk-adjusted improvement (Sharpe goes up significantly)
For our 60% / b=3 example: instead of betting 46.7%, you bet 23.3%. Still substantial, but if you have three losses in a row: (1 - 0.233)³ = 0.451 — 45% remaining vs 15%. Survivable.
Quarter-Kelly for Real Trading
Most professional traders use Quarter-Kelly (12% of bankroll in our example), or even less, for these reasons:
- True win probability is uncertain (your edge estimate could be wrong)
- Correlation between trades increases effective risk
- Tail events (market crashes) aren't in your historical sample
- Behavioral comfort — sleeping at night matters
Applying Kelly to Stock Trading
Stock trading isn't binary like a coin flip, but you can still apply Kelly using estimated win rate and average win/loss size:
Step 1: Estimate Your Edge
Look at your last 20+ trades. Compute:
- Win rate (p): % of trades that were profitable
- Average win ($w): average dollar gain on winners
- Average loss ($l): average dollar loss on losers (positive number)
- Payoff ratio (b) = w / l
Step 2: Compute Kelly
Plug into f* = (bp - q) / b. If f* < 0, your edge is negative — stop trading this setup.
Step 3: Use Quarter-Kelly
Divide f* by 4. This is your position size as a fraction of trading capital.
Worked Example: NVDA Swing Trade
You've taken 30 NVDA swing trades over the past year:
- 17 winners (57% win rate) averaging $850 gain each
- 13 losers averaging $400 loss each
- b = 850 / 400 = 2.125
Kelly: f* = (2.125 × 0.57 − 0.43) / 2.125 = 0.78 / 2.125 = 0.367
Full Kelly: 36.7% of capital. Half Kelly: 18.4%. Quarter Kelly: 9.2%.
With $50,000 trading capital, your next NVDA trade should risk no more than $4,600 at Quarter Kelly — a substantial but psychologically sustainable size.
When NOT to Use Kelly
- You don't have enough historical trades (need 30+ minimum) for reliable win rate
- Your strategy is leveraged or uses options (loss can exceed bet amount)
- You can't tolerate the drawdowns Kelly mathematically produces
- Correlation between positions makes "single bet" assumption invalid
- Tail risk is asymmetric (selling out-of-money options, for example)
The Buffett Connection
Warren Buffett famously concentrates his portfolio in a few high-conviction positions. At one point, American Express represented ~40% of Berkshire's equity portfolio. To outside observers this looked recklessly aggressive. To Buffett — who had estimated the win probability and payoff carefully — it was approximately fractional Kelly sizing on an obvious edge.
The lesson isn't "concentrate everything." It's that position size should be proportional to edge × confidence in that edge. Tiny positions on huge convictions waste opportunity; massive positions on weak signals court ruin.
Common Mistakes
- Estimating win rate from too few trades. Anything under 30 is statistically meaningless. Your "60% win rate" might really be 45% — Kelly would have you bet aggressively into negative expectation.
- Treating each trade as independent when they're correlated. Two long tech stocks aren't two bets — they're one bet on tech. Effective Kelly is much smaller.
- Ignoring transaction costs. Costs erode b directly. A trade with $100 win and $80 loss after costs is very different from $120 / $60 gross.
- Confusing "stop loss" with "loss size." Your stop is a maximum loss. Your average loss (for Kelly) is the actual mean, usually less than max.
Try It in Trade Sizer
10X Rock's Trade Sizer tool implements Kelly sizing for you. Enter your symbol, capital, risk percentage, and the tool computes:
- ATR-based stop-loss distance
- Position size to keep risk within your percentage
- Kelly sizing recommendation (Quarter Kelly by default)
- 1R / 2R / 3R take-profit targets
The Bottom Line
Position sizing is the most underrated skill in trading. A mediocre strategy with great sizing beats a great strategy with poor sizing — every time. Kelly gives you the mathematical foundation. Fractional Kelly (Half or Quarter) gives you the practical wisdom. Use both.
References
- Kelly, J. L. (1956). "A New Interpretation of Information Rate." Bell System Technical Journal.
- Thorp, E. O. (2006). "The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market." Handbook of Asset and Liability Management.
- Poundstone, W. (2005). Fortune's Formula. Hill and Wang.
Disclaimer: The Kelly Criterion is a mathematical tool, not investment advice. Real-world trading involves uncertainties not captured in the formula. Always validate your edge estimates with out-of-sample data and never risk capital you cannot afford to lose.