Why did we test volume confirmation?
We tested it because volume confirmation is close to universal advice, and universal advice is exactly the kind that nobody re-checks. A breakout on heavy volume is supposed to be more reliable than one on thin volume; the claim appears in most technical education material and inside a great many published strategies.
It is also straightforward to test cleanly, which is rarer than it sounds. Take a setup, split its historical occurrences by whether volume was above or below its recent average at the moment of the signal, and compare what happened next. No parameters to tune, no room for the test itself to be fitted.
And it mattered commercially. A volume filter is a natural feature, customers expect one, and building it would have been a day's work. We wanted to know whether shipping it would help anyone.
What did we find?
We found no usable separation. Across a large sample the above-volume cohort did not outperform the below-volume cohort by enough to be distinguishable from noise, and the direction of the small differences that did appear was not consistent across markets.
Inconsistency across instruments is the part that settles it. A real effect driven by participation should show up in the same direction on most liquid pairs, even if the magnitude varies. What we saw looked like what you would expect from splitting any dataset on an uninformative variable.
There is an important caveat we state plainly, because it genuinely limits the finding. Spot forex has no centralised volume. What retail platforms label volume is tick count from a single broker's feed — a proxy for activity, not a measure of traded size. It is entirely possible that genuine volume, if it existed for spot FX, carries information this proxy does not.
But the proxy is what a retail platform actually has. So the practical question is not whether volume matters in principle; it is whether filtering on the number a trader can actually see improves outcomes. In our data, it did not.
What did we do with the result?
We did not ship the feature. QuantParadox does not offer volume as a setup filter or as a sizing input, and we disclose thin volume coverage wherever the underlying data is sparse rather than letting a strategy silently depend on it.
Publishing a negative result about a feature we could have sold is uncomfortable, which is roughly the point. A platform whose entire claim is honest measurement has to be willing to measure things that go against it, and to say so where customers can read it.
The alternative — shipping the filter because customers expect it, knowing it does nothing — is how tools accumulate features that make backtests look busier without making them more informative. Every such feature is another dimension in which to overfit, sold as rigour.
How should you test a filter like this yourself?
Test a filter by splitting the same setup's historical occurrences on the filter condition and comparing what happened next — not by adding the filter and seeing whether the strategy's total improves.
That distinction matters more than it sounds. If you add a volume requirement and the equity curve improves, you have changed two things at once: the filter, and the number of trades. Fewer trades means a noisier result, and a noisier result improves roughly half the time by chance. The comparison you actually want holds the setup constant and asks whether the filtered subset behaved differently from the unfiltered one.
Then check consistency across instruments before believing anything. A real effect driven by a market mechanism should point the same way on most liquid pairs, even if the magnitude varies. An effect that appears strongly on two pairs and reverses on three is what splitting on an uninformative variable looks like, and it will look exactly like a discovery on whichever pair you examined first.
Finally, count the filter variants you tried. Above-average volume, twice-average volume, volume above its 20-period mean, above its 50-period mean — each threshold is a trial, and the best of eight thresholds needs a higher bar than the first one you tested.
Does volume matter anywhere in trading?
Volume matters considerably in centralised markets, and that is precisely why the spot forex case is different rather than an argument against volume generally.
Futures, equities and other exchange-traded instruments report actual traded size through a central venue. That number reflects real transactions by real participants, and a large body of research supports its informational content. Volume analysis there rests on something concrete.
Spot forex has no central exchange, so no consolidated volume exists. Every platform showing you a volume histogram on EUR/USD is showing you tick counts from one broker's feed — how often the price updated, not how much was traded. Those two things correlate, but the correlation is not tight enough to inherit the research.
If you trade FX futures rather than spot, the volume figures are real and the conventional analysis applies. If you trade spot, the number on your chart is not measuring what the textbook assumed.