We backtested our own signal — here is what failed

Most market tools never tell you whether their signal actually works. That is not usually because they are lying. It is because nobody checked, and checking is the kind of thing that can ruin a good story.

We checked. Here is what we found, including the part we did not want.

The setup

We took a full year of daily price and volume data for 50 large stocks. For every stock, on every day, we calculated how unusual that day's trading volume was compared with that stock's own previous 20 days. Then we looked at what the price did over the following 1, 5 and 20 trading days.

That is roughly 10,500 separate observations. The question was simple: after an unusual-volume day, does the price go up more than normal?

The answer was no

Across all days, the average 5-day forward return was about +0.54 percent. That is the baseline: what you would get from picking a random day.

On days when volume was at least twice its normal level, the average 5-day return was −0.19 percent. Slightly worse than doing nothing.

We tried the obvious refinement: what about unusual volume on a day when the price also closed higher, the classic "volume confirms the move" idea? Average 5-day return: +0.49 percent. Essentially identical to the baseline. No edge.

So the appealing story, that heavy volume means smart money is accumulating and you should follow, did not survive contact with the data.

Then we asked a different question

Instead of asking which way the price went, we asked how far it moved, ignoring direction entirely.

Baseline absolute 5-day move: 4.40 percent. After a volume z-score of 2 or higher: 5.26 percent. That is roughly 18 percent more movement, on a sample of over 800 occurrences.

That result held up across variations. Unusual volume does tell you something real. It tells you the asset is more likely to move meaningfully in the near future. It just refuses to tell you which way.

What we did about it

We had two options. Bury the finding and market the product as a signal service, which would have been easier and is what a lot of tools do. Or redesign around what the data actually supports.

We redesigned. Momentu is built as an activity radar, not a buy signal. The unusual-volume score answers "is something happening here." A separate trend column answers "which way has this been going." We keep them apart on purpose, because merging them would imply a relationship the evidence does not support.

The limits of our own test

Honesty cuts both ways, so here is where our study is weak.

It covers one year, and that year had an upward drift, which inflates the baseline everything is measured against. The 50 stocks are ones we already track, so there is survivorship bias. We ignored transaction costs. And the forward return windows overlap, which makes the effective sample smaller than 10,500 independent observations.

None of that overturns the core finding, which was consistent across every cut we tried. But anyone telling you their backtest is airtight is telling you something else about themselves.

Why publish this

Because the alternative is asking you to trust a number we never verified. And because the finding is genuinely useful once you accept it: knowing where movement is likely, without pretending to know its direction, is a real and modest edge in attention. It is not a crystal ball. We would rather be the tool that tells you that plainly.

Momentu's free daily radar shows unusual activity across 60 assets, with trend displayed separately, so you always see both facts without either masquerading as the other.

Momentu is an informational tool, not financial advice or a buy/sell recommendation, and it does not consider your personal circumstances. Any decision you make is your own.