VAQO ML — A Real Machine-Learning Model for Stocks

A machine-learning model ranking stocks by probability to rise and risk — fully transparent about training data, accuracy and limits. Every prediction runs in your browser.

What the model actually does

Two statistical models run inside your browser, with nothing sent to a server. They take a price series and return three separate scores: momentum, risk and opportunity.

They do not forecast a price and do not recommend an action. They measure behaviour: how strong the trend is, how volatile the asset is, and how those two relate.

Accuracy, including the unflattering figure

The direction model was trained on 73,781 samples and reaches 54.7% accuracy over a 10-day horizon. Said plainly: that is barely better than a coin flip. Anyone promising accurate directional forecasting in equities is selling something else.

The volatility model was trained on 120,467 samples and reaches 77.1%. The gap makes sense: predicting how much an asset will move is far easier than predicting which way. In practice the risk measure is the useful half.

That gap is why the system shows no single arrow. Three separate measures let you see that a stock is strong on momentum and also risky — information an average would erase.

What the model does not know

The model sees price behaviour and nothing else. It does not read filings, does not know about a pending merger, and does not understand a regulatory change. An event absent from the price series simply does not exist to it.

Weights retrain automatically each week on current data. If accuracy falls, the figure on this page falls with it — we do not freeze a number that was once true.

Last updated: 2026-09-27