The Kelly Criterion Explained — How Math Finds Your Optimal Position Size
Most traders obsess over what to buy. The truly successful ones obsess over how much. Position sizing is arguably more important than stock selection — because even the best signal in the world will destroy your account if you bet too big, and will barely move the needle if you bet too small.
The mathematical answer to "how much should I bet?" has been known since 1956. It's called the Kelly Criterion — and it's used by professional gamblers, hedge funds, and anyone serious about optimizing growth under uncertainty. Let's understand it.
The Formula
The Kelly Criterion is beautifully simple:
Where:
• f* = fraction of your bankroll to bet
• b = net odds received on the bet (e.g., if you bet $1 to win $2, b = 2)
• p = probability of winning
• q = probability of losing (1 − p)
In plain English: bet more when you have a bigger edge, and less when your edge is small. Never bet when you have no edge at all.
A Simple Example
- The Kelly formula: f* = (bp - q) / b where b = odds, p = win probability, q = loss probability
- A strategy with 60% win rate and 1:1 risk-reward yields an optimal bet of 20% of bankroll
- Most professional traders use ½ Kelly or ¼ Kelly to reduce volatility while keeping most of the growth
- Source: J.L. Kelly Jr., "A New Interpretation of Information Rate", Bell System Technical Journal, 1956
You're offered a bet: flip a coin. If it's heads, you win $2 for every $1 bet. If tails, you lose your $1. The coin is fair (p = 0.5).
- b = 2 (you win $2 for every $1 risked)
- p = 0.5 (50% chance of winning)
- q = 0.5 (50% chance of losing)
f* = (2 × 0.5 − 0.5) / 2 = (1.0 − 0.5) / 2 = 0.5 / 2 = 0.25
Kelly says: bet 25% of your bankroll on each flip.
This maximizes your long-term growth rate. Bet more than 25%, and the volatility eats into your compounding. Bet less, and you leave growth on the table.
Kelly in Trading: Adapting the Formula
Trading isn't a binary coin flip with fixed odds. In trading, we have:
- A signal's probability score (p) — from backtesting or model output
- A risk-reward ratio (b) — where b = (target profit) / (stop-loss distance)
So for a trade with:
- Target profit: $3 per share
- Stop-loss: $1.50 per share
- Probability score: 60%
b = 3 / 1.5 = 2
f* = (2 × 0.6 − 0.4) / 2 = (1.2 − 0.4) / 2 = 0.8 / 2 = 0.40
Kelly says: allocate 40% of your capital to this trade. That's insane for real-world trading. Which brings us to the most important concept in applied Kelly theory.
The Fractional Kelly — Why You Should NEVER Use Full Kelly
Full Kelly maximizes long-term growth rate, but it comes with brutal drawdowns. A 40% allocation on a single trade means a string of 3 losses wipes out 78% of your capital. That's mathematically optimal in theory — but psychologically devastating in practice.
The solution: Fractional Kelly.
• Full Kelly = maximum theoretical growth rate, but 50%+ drawdowns are common
• Half Kelly (0.5 × f*) = ~75% of the growth rate with ~25% of the volatility
• Quarter Kelly (0.25 × f*) = ~50% of the growth rate with dramatically lower drawdowns
Most professional traders use ¼ Kelly or ½ Kelly. The smoother equity curve is worth the slightly lower theoretical returns.
At quarter Kelly, our 40% allocation becomes 10%. At half Kelly, it becomes 20%. Both are far more realistic for a human trader managing real money.
How GemStox Applies Kelly
Every GemStox signal includes a probability score derived from cross-validation across multiple strategy classes. That probability score feeds directly into Kelly-based position sizing:
1. Calculate raw Kelly fraction: f* = (b × p − q) / b
2. Apply half-Kelly multiplier: f_half = f* × 0.5
3. Cap at 5% max position size (risk management override)
4. Output: "Suggested position: 3.2% of portfolio with stop at $X"
This means you're not just getting a "BUY" signal — you're getting a mathematically optimized position size that accounts for the signal's probability, your risk-reward setup, and real-world drawdown tolerance.
Why Position Sizing Matters More Than Stock Picking
Here's a thought experiment: two traders both start with $100,000 and trade the exact same signals for a year.
- Trader A uses fixed 2% position sizing on every trade, regardless of signal strength.
- Trader B uses Kelly-based sizing: 1% on low-conviction signals, 4% on high-conviction signals.
At the end of the year, with identical trade selection, Trader B outperforms Trader A by 40–60% — because they bet more when the odds were in their favor and less when they weren't.
Position sizing is leverage on your edge. Kelly tells you exactly how much leverage to use.
Get probability-based signals with built-in position sizing.
Every GemStox signal includes a probability score and suggested Kelly-optimal position size. Start with a $7 Day Pass.
See Plans & Pricing →