A formula built in a Bell Labs research paper about noisy telephone lines became, decades later, the mathematical backbone of how professional gamblers and some of the most successful hedge funds in history size every single bet.
The Question Every Trader Answers Badly
Ask most traders how they decide position size and you'll get an answer like "I usually put in a few thousand dollars" or "I go bigger when I feel more confident." Compare that to how they usually treat entries - watching charts, waiting for confirmation, checking multiple indicators - and there's an obvious mismatch. Most traders spend 90% of their preparation deciding whether to take a trade and 10% deciding how much to risk on it, when the sizing decision is arguably more responsible for long-term results than the entry itself.
The Kelly Criterion exists to fix exactly this gap. It's a formula that answers, with actual math instead of a feeling, what fraction of your capital you should risk on a given trade based on two things you can measure: how often you win, and how much you win relative to how much you lose.
Where It Came From
In 1956, a Bell Labs scientist named John Kelly Jr. published a paper about a completely different problem - how to maximize the rate of information transmitted over a noisy phone line. Buried inside the math was a formula for optimal resource allocation under uncertainty that turned out to generalize far beyond telecommunications.
Professional gamblers picked it up first, using it to size blackjack bets based on a card counter's shifting edge. The person most responsible for bringing it into investing is Ed Thorp - a mathematician who wrote the book on beating blackjack, then applied the same Kelly-based sizing logic to warrant pricing and eventually ran one of the most successful quantitative hedge funds of his era. The throughline from a 1956 signal-processing paper to modern systematic trading desks runs almost entirely through Kelly's original insight: when you have a genuine, measurable edge, there is a mathematically optimal amount to bet on it - and it's neither "as much as possible" nor "a fixed amount regardless of the edge."
The Formula
The full Kelly formula in its gambling form is:
f = (bp - q) / b*
Where f* is the fraction of capital to wager, b is the net odds received, p is your win probability, and q is your loss probability (1 - p).
For trading, it's more useful in this simplified form:
f = W - [(1 - W) / R]*
- W = your historical win rate, as a decimal (a 55% win rate is 0.55)
- R = your win/loss ratio - your average winning trade's size divided by your average losing trade's size
- f* = the fraction of your total capital to risk on the next trade
A Worked Example
Say your trading journal shows the following over your last 50 trades:
- Win rate (W): 55% (0.55)
- Average winning trade: +9%
- Average losing trade: -6%
- Win/loss ratio (R): 9 ÷ 6 = 1.5
Plugging into the formula:
f* = 0.55 - [(1 - 0.55) / 1.5] f* = 0.55 - [0.45 / 1.5] f* = 0.55 - 0.30 f* = 0.25
Full Kelly says to risk 25% of your total capital on this trade. To most traders with any real market experience, that number should immediately look alarming - and that instinct is correct.
Why Nobody Actually Uses Full Kelly
Full Kelly is mathematically optimal for maximizing long-run compound growth if your win rate and win/loss ratio are exactly correct and never change. Real trading violates both assumptions constantly:
Your edge is an estimate, not a fact. A 55% win rate calculated from 50 trades has real statistical noise around it - your true long-run win rate could easily be 48% or 62%. Full Kelly sizing punishes an overestimated edge severely, because you're betting a large fraction of capital on a number that might be wrong.
Your edge changes over time. Market regimes shift. A strategy with a real edge in a trending market can lose that edge entirely in a choppy, range-bound one. Full Kelly has no mechanism for "my edge might currently be lower than my historical average" - it just uses whatever number you feed it.
The drawdowns are genuinely brutal. Even when the underlying edge is completely real, full Kelly betting produces enormous swings in account value. Simulated full-Kelly betting sequences regularly show 50%+ peak-to-trough drawdowns even with a real, sustained edge, purely from the mathematics of variance. Very few traders can stay disciplined through a 50% drawdown without abandoning the strategy at exactly the wrong moment - which defeats the entire purpose of using a rigorous formula in the first place.
The Practitioner's Fix: Fractional Kelly
This is why virtually every professional application of Kelly sizing uses a fraction of the full recommendation - most commonly half Kelly, sometimes quarter Kelly for traders who want a wider margin of safety.
| Sizing approach | Capital risked (from example above) | Growth rate vs. full Kelly | Drawdown vs. full Kelly |
|---|---|---|---|
| Full Kelly | 25% | 100% (baseline) | 100% (baseline) |
| Half Kelly | 12.5% | ~75% of full Kelly growth | ~50% of full Kelly drawdown |
| Quarter Kelly | 6.25% | ~44% of full Kelly growth | ~25% of full Kelly drawdown |
The relationship isn't linear, and that asymmetry is the entire argument for fractional Kelly: cutting your bet size in half only costs you about a quarter of your theoretical growth rate, but it cuts your expected drawdown roughly in half. Going further to quarter Kelly gives up less than half the growth rate for a quarter of the drawdown risk. For anyone whose edge estimate carries real uncertainty - which is everyone - that trade-off is close to free insurance against being wrong about your own numbers.
The Part Most People Skip: You Need a Real Number for W and R
The formula is only as good as its inputs, and this is where most people's Kelly calculation quietly falls apart. "I win most of the time" is not a win rate. "My winners feel bigger than my losers" is not a win/loss ratio.
To get real inputs, you need a trading journal with at minimum 30-50 closed trades logged with entry price, exit price, and result. From that log:
- Win rate (W) — the percentage of logged trades that closed profitably
- Average win — the mean percentage gain across only the winning trades
- Average loss — the mean percentage loss across only the losing trades
- R — average win divided by average loss
Without this, any Kelly number you calculate is really just a guess dressed up in a formula - which is precisely the false precision the formula's own logic exists to warn against.
Where Alerts Fit Into This
Kelly sizing answers how much to risk. It says nothing about whether you'll actually exit at the loss size your calculation assumed. This is the quiet failure point in most position-sizing plans: a trader correctly sizes a position at, say, half-Kelly assuming a 6% average loss, then doesn't exit when the trade hits -6% because "it'll probably come back" - turning a planned 6% loss into an unplanned 15% one.
That single behavior invalidates the entire calculation. Your R ratio was built from your historical average loss size; if your actual losses run larger than that average because you hesitate to exit, your real edge is worse than the number you plugged into the formula.
This is exactly what a stop-loss price alert is for - not a trading signal, but an enforcement mechanism. Set the alert at the exact loss level your position sizing assumed, and let it notify you the moment price reaches it, so the exit happens on schedule instead of being renegotiated in real time by hope.
price falls 6% below entrySet a stop-loss alert at the exact loss threshold your position sizing calculation assumed - so your real win/loss ratio matches the one you calculated it to be.
Enforce the exits your position sizing depends on
Stock Alarm Pro lets you set price, percentage-move, and technical alerts on every position so your actual losses match the ones your risk plan assumed.
S&P 500 Screener
Filter by metrics, fundamentals
Price Alerts
Never miss a move
35+ Global Markets
Stocks, crypto, futures
AI Analysis
Ask questions, get answers
Practical Takeaways
- Calculate your real W and R from a trading journal, not from memory or gut feel — at least 30-50 trades.
- Never use full Kelly. Start at half Kelly, and consider quarter Kelly if your edge estimate is based on a small sample or a strategy you've traded for less than a year.
- Re-calculate periodically. Your win rate and win/loss ratio drift as market conditions change and as your strategy evolves - a Kelly calculation from two years ago may no longer reflect your current edge.
- Treat a negative Kelly result as real information. If W - [(1-W)/R] comes out negative or near zero, the math is telling you that, based on your own logged results, you don't have a demonstrable edge in this strategy yet - which is worth knowing before risking more capital on it, not after.
- Size the bet, then enforce the exit. The formula only protects you if your real losses match the ones it was calculated from - which means the stop-loss alert is not optional, it's the other half of the system.
The Bottom Line
The Kelly Criterion won't tell you which stocks to buy or when to buy them. What it does is replace a guess - "how much should I put into this?" - with a number derived from your own measured track record. Used at full strength it's too aggressive for almost anyone; used at half or quarter strength, with honest inputs from an actual trading journal, it's one of the few pieces of trading math that translates directly into a smaller, calmer answer to a question most traders currently answer with a feeling.
