Kelly Criterion: Let Your Own Data Tell You How Much to Risk Per Trade

"Never risk more than 2%." It's the rule everyone repeats. Nobody knows where it came from. It doesn't know your win rate. It doesn't know how big your average winner is relative to your loser. It doesn't know anything about your system — it just sounds responsible, so it spread.

The Kelly Criterion, originally developed by John Kelly Jr. at Bell Labs in 1956, does the opposite. It takes your actual measured edge — your win rate and your payoff ratio — and outputs the position size that maximizes long-run account growth. Not a universal rule. A formula derived from your own trade data.

There's a catch, and it's an important one: Kelly is only as good as the inputs you feed it. If your win rate estimate is off, or your R-ratio is calculated from too small a sample, Kelly will oversize you — and oversizing is where accounts die. This article covers the formula, how to interpret it, and the data requirements you need to meet before trusting the output.

The Kelly Criterion Formula (Simplified for Traders)

The full Kelly formula handles multi-outcome scenarios. For trading — where you're essentially dealing with wins and losses at varying magnitudes — the simplified version is:

Where:

• W = your win rate as a decimal (e.g. 45% = 0.45)
• R = your win/loss ratio: average winner size divided by average loser size

The output is the percentage of your account capital to risk on a single trade. A Kelly output of 0.20 means risk 20% of your account. Most traders immediately cut that in half — more on why below.

If Kelly outputs a negative number, it means your system has negative expectancy: you'd be better off not trading it at all. That's a useful signal in itself. Chasing win rate without tracking your payoff ratio is why so many traders never notice this until too late.

A Worked Example

Suppose your last 80 trades show:

• Win rate: 45% → W = 0.45
• Average winner: $250 | Average loser: $100 → R = 2.5

Full Kelly says risk 23% of your account per trade to maximize compounded growth given this edge. That's a number that will make most traders uncomfortable — and it should. The math is sound, but it assumes your inputs are perfectly accurate. Because they're almost certainly not, you apply a safety margin. Half-Kelly brings you to 11.5%, which still exceeds what most traders risk but is far less likely to create catastrophic drawdowns when the inevitable estimation error appears.

Why Most Traders Use Half-Kelly

Kelly is derived from an idealized model. It assumes your win rate and R-ratio are stable, accurately measured, and will continue into the future. None of those assumptions hold perfectly in live trading.

The problem is asymmetric. If Kelly tells you 20% and you're slightly wrong about your edge, you can lose 30-40% of your account in a normal losing streak. But if you use half-Kelly and your edge is exactly as estimated, you only give up a fraction of theoretically maximum growth. The downside risk of full Kelly far exceeds the upside benefit over full-Kelly — which is why serious traders almost universally use a fractional Kelly: most commonly 0.5× (half-Kelly) or 0.25× (quarter-Kelly).

"The Kelly Criterion maximizes the geometric mean of wealth — but only if your inputs are exact. In practice, fractional Kelly is the rational choice."

The more uncertainty in your input estimates, the lower the fraction you should apply. A trader 6 months into live trading with 60 trades logged should probably use quarter-Kelly at most. A trader with 500 trades across stable market conditions has more confidence in the stability of their parameters and might go to half-Kelly.

The Data Requirements: When Can You Actually Trust Kelly?

This is the part most articles skip. Kelly is a function of two numbers: win rate and R-ratio. Both need to be stable estimates from a large enough sample before they mean anything.

Beyond trade count, the quality of the data matters as much as the quantity. Trades need to be taken with consistent stop placement — if your stop size varies wildly, your R-ratio becomes noise. Kelly also assumes your strategy parameters are stable: if your win rate was 55% in a trending market and you're now running Kelly on that figure in a choppy market, you're trading on stale inputs.

This is why monitoring your system for edge decay matters before applying Kelly. If your edge is changing, your Kelly percentage should change with it.

Kelly vs. Fixed-R: Which Should You Use?

They're not mutually exclusive — Fixed-R sizing is how you normalize every trade to a consistent risk unit so that your journal data becomes meaningful. Kelly is what you calculate from that journal data to determine what percentage your Fixed-R should be.

The practical workflow looks like this: start with a fixed risk percentage (1-2%) when you don't yet have enough data. Log every trade consistently with a well-defined stop. After 50-100 trades, calculate your win rate and R-ratio from the clean data. Feed those into Kelly, apply your chosen fraction, and adjust your risk percentage accordingly. Then keep monitoring — if your statistics drift, Kelly will tell you to resize.

Fixed-R without Kelly is arbitrary. Kelly without Fixed-R is unmeasurable. Together, they form a complete position-sizing framework grounded in your actual performance.

How Kelly Connects to SQN and Expectancy

Kelly, SQN, and expectancy are three lenses on the same underlying question: does your system have a real edge, and if so, how much should you bet on it?

Expectancy tells you the average R-multiple per trade — your expected profit for every unit of risk taken. A positive expectancy system is one worth trading. Expectancy is an input to Kelly: your R-ratio and win rate are just a decomposition of expectancy into its two components.

SQN (System Quality Number) adds a third dimension — consistency. Two systems can have identical expectancy but very different SQN scores. The one with lower variance in outcomes has a higher SQN and a more reliable Kelly estimate. A high-SQN system is one where you can trust the Kelly output more, because the edge isn't just positive but stable.

In practice: calculate expectancy to confirm you have an edge, use SQN to confirm that edge is consistent and reliable, and then apply Kelly (fractionally) to determine the position size that maximizes your compounded growth. Each metric is a prerequisite for the next.

How SignalDeck Calculates Kelly Automatically

SignalDeck computes Kelly Criterion, half-Kelly, expectancy, SQN, and R-multiple in real-time from your closed trade journal — no spreadsheets, no manual inputs. As you log trades, your Kelly percentage updates automatically and is visible alongside your other system metrics.

You can filter by strategy, tag, date range, or asset class to see how your Kelly percentage differs across different setups. If your "Breakout" strategy has a Kelly of 18% and your "Counter-trend" strategy shows 3%, you have a clear, data-driven basis for how much confidence to place in each — and where to concentrate capital. Try it free during beta.