MEASURED · 12 SEPTEMBER 2026

Does the risk-reward ratio actually work?

We ran every ratio from 1:0.5 to 1:5 across 66,572 gold trades and 2,000,000 minutes of price history. The win rate fell to offset every increase in target size, every time, to within a rounding error. The ratio is not an edge. It is a choice about variance — and about whether you can afford to wait.

The advice, and the arithmetic that undercuts it

“Never take a trade below 1:2” is close to universal. The logic sounds airtight: if your winners are twice the size of your losers, you can be wrong most of the time and still come out ahead. Being right only a third of the time and still profiting is a genuinely attractive idea, and it is the reason the rule spread.

The problem is the half of the sentence nobody says out loud. Moving your target further away does not only make winners bigger, it makes them rarer. Price has to travel further to reach a distant target, and further travel is less likely. A 2:1 ratio needs a 33.3% win rate to break even, and a 1:1 needs 50%. The question that decides everything is whether the win rate falls to exactly that line, or stops short of it, or overshoots. If it lands exactly on the line, the ratio is arithmetic rather than an edge.

That is measurable, and as far as we can tell nobody had published the measurement. So we ran it.

How it was tested

The entries are deliberately worthless. One every 60 minutes, taken long and short at the same instant, with no indicator and no filter. That sounds like a strange way to test anything, and it is the entire point: if a ratio produced an edge by itself, it would have to show up against entries like these, because nothing else in the test could produce one.

Taking both directions at once matters more than it looks. Gold roughly doubled across this sample. A long-only test would have reported that trend as a property of the risk-reward ratio and produced a confident, meaningless result. Opening both sides cancels the drift.

Each trade is a bracket: a stop loss a fixed distance away, a take profit at that distance multiplied by the ratio. Minute bars are walked forward and whichever level is touched first wins. Three details do most of the work in keeping the answer honest:

• Ambiguous bars count as losses. When one minute covers both the stop and the target, which came first is unknowable from the bar. Resolving those as wins would let us manufacture an edge out of nothing, so they are counted as losses and reported as a column.
• The spread is the real one. Not a flat assumption — the per-bar spread the broker actually recorded at entry, median $0.200 per ounce.
• Unresolved trades are reported, not hidden. Anything still open when the holding window expires gets its own column. This turned out to matter far more than expected.

The result

A 5 dollar stop, 66,572 trades at every ratio, 2021-01-15 to 2026-09-11. “Break-even” is the win rate the ratio needs with zero costs, which is simply 100/(1+RR). “Gap” is what we measured minus that.

Read the gap column. At 1:1 the market delivered 49.91% where the arithmetic demanded 50%. At 2:1, 33.44% against 33.33%. At 3:1, 25.15% against 25%. Seven of the eight gaps are smaller than their own 95% confidence interval, which is the statistical way of saying they are indistinguishable from zero. Pooling every ratio and every stop distance that resolved cleanly — 17 rows in all — the average gap is -0.021 percentage points.

The win rate does not approximately offset the target size. It offsets it almost exactly, across five and a half years and more than half a million trades.

So what is the last column doing there?

Every ratio lost money, and it lost roughly the same amount: about twenty cents per ounce. Compare that to the median spread of $0.200. At 1:1, expectancy was -$0.2165 against a spread of $0.200.

That is the whole story in one comparison. With no edge at entry, what you lose is what you pay. The ratio does not change the cost of a trade — it changes how many trades you take to cover the same ground, and every one of them pays the spread again. That is why costs, not ratios, are where the leverage is, and why we have separately measured how the gold spread moves by hour.

The finding we nearly published backwards

Widening the stop to $20 appeared to break everything. High ratios fell apart: a 5:1 target won just 6.99% of the time against a 16.67% break-even. Taken at face value, that is a headline — wide stops punish ambitious targets.

It is not a market effect. It is the holding window. 39.63% of those trades never touched either level before time ran out, and the ones cut off are disproportionately the slow winners — a distant target takes longer to reach, so it is the first casualty of a deadline. Across all 32 rows in the study, the correlation between the unresolved rate and the gap from theory is -0.97. Where trades were actually allowed to finish, the gap vanishes.

We are describing this at length because the wrong version of this article is easy to write and impossible to detect from the outside. Drop the unresolved column and you have a confident, well-evidenced, completely false claim that high risk-reward ratios lose money. They do not. Trades that get cut off do.

Which makes the artifact a real warning in its own right. If anything forces you to close early — a session rule, a prop firm daily reset, a habit of flattening before the weekend — distant targets are the first thing it takes from you, and you will experience it as the ratio failing rather than the deadline.

What this changes

Nothing about whether to use a stop and a target. Everything about what you expect them to do.

A risk-reward ratio is not a filter that turns a bad entry into a good one, and no amount of raising it will rescue an entry with no edge. What it does control is the shape of your results: how often you win, how long you wait, how long the losing runs get. Those are real decisions with real consequences — targeting five times your risk wins 16% of the time, and that produces losing streaksmost people abandon before the maths gets a chance to work. Just do not mistake any of it for an edge.

Our own bot fixes the ratio at 2.0 in every market regime. An earlier version scaled it by regime, and validating that on out-of-sample data showed it was fitting noise. Removing the feature improved results. This study is the general case of that specific finding, and you can work through the arithmetic for your own account with the risk-reward calculator or see how the ratio is applied in the gold trading bot itself.

Limits

One symbol, one broker, 2021-01-15 to 2026-09-11. Gold is unusually trend-prone and unusually volatile, and the result may not transfer to instruments that behave differently. The entries are random by design, which is what isolates the ratio but also means this says nothing about how a ratio interacts with an entry that genuinely has an edge — plausibly the more useful question, and a harder one, because the answer depends on the entry.

Fills are assumed at the touched level with no slippage beyond the spread, which flatters every row roughly equally. And the holding-window effect described above means the 4:1 and 5:1 rows at wider stops should be read as incomplete rather than as results. The raw numbers for all four stop distances are published alongside our other studies, including the columns that make this argument falsifiable.

Common questions