The Kelly Criterion: How Much Should a Systematic Trader Risk Per Trade?

A backtest can tell you that a strategy has an edge. It cannot tell you how much to risk on the next trade. That gap, between "I have an edge" and "here is my position size", is where a surprising number of otherwise-good systems die, and it is exactly the gap the Kelly criterion was built to close.

As a software engineer approaching trading systematically, I find Kelly one of the most clarifying pieces of math in the whole field. Here is the plain-language version.

What problem does the Kelly criterion solve?

Suppose your tested strategy wins 55% of the time and the average win equals the average loss. You have a real edge. Now: what fraction of your account do you risk per trade?

Risk too little and you leave growth on the table. Risk too much and, this is the counterintuitive part, you can turn a winning strategy into a losing account. Position sizing is not a detail; it is a variable that can flip the sign of your outcome.

The Kelly criterion, published by John Kelly at Bell Labs in 1956, answers the question directly: it gives the bet fraction that maximizes the long-run compound growth rate of your capital.

How is the Kelly fraction calculated?

For a simple win/lose bet the formula is:

f* = W − (1 − W) / R

where W is the win rate and R is the ratio of average win to average loss.

A textbook illustration (not a result from any KenKem system): with a 55% win rate and wins equal to losses (R = 1), Kelly says f* = 0.55 − 0.45/1 = 10% of capital per trade. With a 40% win rate but wins twice the size of losses (R = 2), f* = 0.40 − 0.60/2 = 10% as well. Two very different-feeling strategies, same optimal fraction, the formula sees through the surface to the underlying edge.

More generally, the Kelly fraction is approximately edge divided by variance, which is the version that matters for real strategies with messy trade distributions.

If the formula returns zero or negative, Kelly is telling you something important: you have no edge, and the optimal bet is nothing.

What happens if you bet more than Kelly?

This is the part every trader should internalize, because it is not intuition, it is arithmetic.

Kelly is a peak, not a floor. Bet beyond the peak and your long-run growth rate goes down, even though every individual trade still has positive expectancy. At exactly 2× Kelly, the expected growth rate of your capital is zero. Beyond 2×, a strategy that wins on average will still grind your account toward ruin, because the losses compound faster than the wins can repair them.

The mechanism is the asymmetry of percentages: a 50% drawdown needs a 100% gain to break even. Oversized positions dig holes faster than any edge can climb out of them.

In other words: overbetting a winning strategy is a mathematically guaranteed way to lose. No amount of signal quality rescues a broken sizing rule. This is why I treat position sizing as a first-class engineering problem, not an afterthought.

Why does almost nobody trade full Kelly?

Full Kelly assumes you know your true win rate and payoff ratio. You never do. You have backtest estimates, noisy samples from a market that also changes regimes underneath you.

The danger is asymmetric: if your real edge is smaller than your backtest suggested (it usually is), full-Kelly sizing based on the optimistic estimate puts you past the true peak, into the region where growth degrades or turns negative. Overestimating your edge while betting full Kelly is a silent account-killer.

So practitioners run fractional Kelly, typically one-quarter to one-half:

• Half Kelly gives up roughly 25% of the theoretical growth rate while cutting the size of drawdowns roughly in half.
• Quarter Kelly is more conservative still, commonly used when the edge estimate itself is uncertain, which is most of the time.

That trade, a modest slice of growth in exchange for a dramatically more survivable equity curve, is one of the best deals in trading.

How do Kelly and drawdown connect?

Kelly sizing and drawdown limits are two views of the same decision. Every Kelly fraction implies a distribution of drawdowns: full Kelly implies stomach-churning ones (deep drawdowns are not a tail risk at full Kelly, they are expected behavior), while fractional Kelly compresses them sharply.

Run the logic in reverse and it becomes practical: start from the maximum drawdown you (or your prop firm) can tolerate, and let that ceiling dictate your Kelly fraction, not the other way around. This is the same drawdown-first philosophy behind the Calmar ratio: the constraint that matters is the hole you can survive, and the sizing follows from it.

How does this shape the KenKem workflow?

KenKem builds software-assisted trading tools, TradingView indicators and MT5 Expert Advisors for Gold and Crypto, and the Kelly lesson is embedded in their design: risk per trade is the user's most important setting, and it stays in the user's hands. The EA executes according to the risk fraction, daily risk cap, and drawdown limits the user configures; KenKem does not choose the size, manage the account, or promise that any fraction is safe.

To be transparent, as always: our systems do not yet have a live track record, and no formula, Kelly included, turns an uncertain edge into a certainty. What Kelly gives a systematic trader is discipline in the sizing decision: a principled reason to bet smaller than feels exciting, grounded in arithmetic rather than fear or greed. Removing that decision from emotion is the entire reason I trade with software.

Educational purpose only. Not financial advice. The examples above are textbook illustrations, not performance claims, please test carefully and configure any trading software according to your own risk tolerance.