Guide · Risk
The short answer
Between 0.25% and 2% for most systematic traders, and the number follows from drawdown tolerance rather than confidence.
Between 0.5% and 2% of the account per trade is the conventional band, and the conventional band is roughly right — but for a reason worth understanding rather than a rule worth obeying.
The reason is that losing streaks are longer than intuition suggests, and the damage from one compounds into the size of the next.
Risking a fixed fraction, each loss shrinks the account, so each subsequent loss is smaller in dollars. That is the point of fixed-fractional sizing: it cannot ruin you outright. What it does do is grind.
| Risk/trade | 5 losses | 10 losses | 20 losses |
|---|---|---|---|
| 0.5% | 97.5% | 95.1% | 90.5% |
| 1% | 95.1% | 90.4% | 81.8% |
| 2% | 90.4% | 81.7% | 66.8% |
| 5% | 77.4% | 59.9% | 35.8% |
| 10% | 59.0% | 34.9% | 12.2% |
At 2% risk, twenty consecutive losses leaves two-thirds of the account — recoverable. At 10% it leaves an eighth, which needs a 719% gain to undo. The difference between the rows is not aggression, it is whether the strategy gets to continue.
For a strategy winning 40% of the time, the chance of ten consecutive losses somewhere in 200 trades is not remote — it is likely. Streaks of that length are an ordinary property of the distribution, not evidence that anything has broken.
Where this goes wrong
This is why the risk figure has to survive the worst streak you have not had yet, not the worst one you have had. Sizing off historical maximum drawdown fits the number to the sample and understates the tail.
The Kelly criterion gives the fraction that maximises long-run growth. For a strategy with edge and payoff it is:
A 55% strategy at 1:1 gives f* = 0.10 — ten percent of the account per trade. Almost nobody should take that, for two reasons.
Half-Kelly gives around three-quarters of the growth with roughly half the drawdown, and quarter-Kelly is where most practitioners land. On the example above, quarter-Kelly is 2.5% — which is how the conventional 1–2% band gets derived rather than assumed.
Varying risk by conviction is intuitive and rarely survives contact with a record. It requires your conviction to be calibrated — that the trades you rate highly genuinely win more often — and almost nobody checks whether theirs is.
Where this goes wrong
Conviction is measured at entry or not at all. Rating a trade after seeing how it went is recording the outcome, not the judgement.
Fixed-fractional sizing scales automatically: risk 1% and the dollar amount rises with the account and falls with it. That is the property that makes ruin arithmetically hard, and it has one consequence people find uncomfortable.
After a drawdown, position sizes are smaller, so the recovery is slower than the fall. That is not a flaw. It is the same mechanism that stopped the fall becoming terminal, and overriding it — trading the pre-drawdown size on a reduced account — raises the effective risk percentage exactly when the evidence for the strategy is weakest.