Somebody asked us this on Reddit in April, and it deserved a better answer than it got:
Do you track what percent of these alerts actually follow through into trend days vs mean-reversion?
We do now. The answer is uncomfortable enough that it is worth writing down properly: our volume alerts have no directional edge whatsoever. They are still useful, but not for the reason most people assume.
How this was measured
Every minute of every stock we track is scored. When a minute prints unusual volume against that stock's own recent baseline, it becomes an alert. The question is what happens next.
The trap in answering that is obvious once you see it: if you only look at alerted minutes, everything looks dramatic, because you selected for drama. You need something to compare against.
So every alerted minute is measured against a control sample: minutes drawn from the same stocks, on the same days, that did not trigger an alert. Same universe, same sessions, same market. The only difference is the alert.
| alerted minutes | control minutes | |
|---|---|---|
| sample size | 1,073,595 | 476,542 |
| higher 15 minutes later | 48.4% | 48.1% |
| higher 60 minutes later | 49.6% | 49.8% |
| average return, 60 min | +0.028% | +0.051% |
| average furthest move up | 1.35% | 1.01% |
| average furthest move down | 1.17% | 0.90% |
Read the first three rows and the alert looks worthless. Read the last two and it looks excellent. Both readings are correct, and that is the entire point.
Direction is a coin flip
49.6% against 49.8%. The alerted minute is fractionally worse than the control, and its average forward return is lower too. Across 22 separate sessions, alerted minutes finished higher more often than control on 13 of them. That is what a coin flip looks like when you count it day by day instead of pooling it.
If you are waiting for the caveat that rescues this, there isn't one. We looked for it.
Range is not a coin flip
Alerted minutes travel 34% further up and 30% further down than control minutes. That held on 22 of 22 sessions, without a single exception.
That is a genuine, large, reliable edge. It is simply not the edge people expect a scanner to have.
More volume means more range, and still no direction
The cleanest evidence is what happens as the volume signal gets stronger. If unusual volume meant "this is going up", then extreme volume should mean "this is really going up".
| relative volume | alerted minutes | higher 60 min later | avg furthest up | avg furthest down |
|---|---|---|---|---|
| 2x to 5x | 318,520 | 49.7% | 1.25% | 1.11% |
| 5x to 10x | 513,107 | 49.8% | 1.35% | 1.17% |
| 10x to 25x | 189,728 | 49.3% | 1.48% | 1.25% |
| 25x and above | 52,240 | 49.2% | 1.59% | 1.35% |
Both range columns climb steadily with volume. The direction column does not move, and if anything it drifts slightly down as volume gets more extreme.
A 25x volume spike is not a stronger buy signal than a 3x one. It is a bigger event, and bigger events are bigger in both directions.
So what is the alert actually good for
Averages are easy to shrug at. The same finding stated as a probability is harder to ignore.
| outcome within 60 minutes | after an alert | control minute | how much more likely |
|---|---|---|---|
| travels at least 3% up | 9.19% | 6.39% | 1.44x |
| travels at least 5% up | 4.57% | 2.78% | 1.64x |
| falls at least 3% | 8.42% | 5.69% | 1.48x |
An alert nearly doubles the chance of a 5% move. That is a real, large, useful edge, and it is the reason a volume scanner is worth watching at all.
Now read the third row. The odds of a 3% drop rise by 1.48x, which is very slightly more than the 1.44x rise in the odds of a 3% gain. The alert is almost perfectly symmetric. It is a genuine forecast that something big is about to happen, and it is completely agnostic about which direction big turns out to be.
That is the honest pitch for a volume scanner, and it is a good one: it finds the 4% of minutes where something is about to happen, out of the 96% where nothing is. It just cannot tell you which way, and no amount of extra volume changes that.
What this actually means if you trade
A volume scanner answers "where is something happening right now". It does not answer "what should I buy". Those are different questions, and conflating them is how people lose money with a perfectly good tool.
Three things follow from the numbers above.
Your entry needs its own thesis. The alert got you to the right stock at the right minute. It contributed nothing to the decision about which way to take it. That decision has to come from somewhere else: the catalyst, the level, the structure, the trend. If your process is "it alerted, so I bought", the data says you have a coin flip with wider outcomes, which is worse than a coin flip after costs.
Size for the range, not for the direction. The one thing the alert reliably predicts is that the next hour will be wider than normal, by roughly a third. That is genuinely actionable, just not as a direction. A position sized for a normal hour is oversized for an alerted one.
Be suspicious of any scanner that only prints the upside. It is easy to publish "these alerts averaged +1.35%" and technically true. We could have. The 1.17% average drawdown is the same statistic viewed from the other side, and leaving it out would turn a range forecast into an implied buy recommendation. Any tool showing you excursions in one direction only is measuring volatility and calling it skill.
The honest limits of this
Two things this does not establish, stated plainly because a study that only lists its strengths is advertising.
The windows are 15 and 60 minutes. The original question asked about 30 and 120 as well. Those are not in the precomputed set, so they are not answered here. The shape is consistent across the two windows we do have, but consistency is not proof it holds at four hours.
This measures alerted minutes, not trades. It says nothing about whether a specific setup built on top of an alert has an edge, because a setup adds entry rules, exits and position sizing that this deliberately strips away. The finding is about the raw signal, which is exactly what was asked about.
What it does establish, on more than one and a half million minutes with a control and a day by day check, is that the volume alert itself is a range forecast. Used as one, it is a good tool. Used as a direction forecast, it is a coin flip that costs you spread.