What exactly does it say?
After each close, every stock with enough valid data for the model gets one word for the next session: "up", "down" or "stable". A stock whose history is too short, or whose data has a break that prevents computing the inputs, gets no forecast for that session. The word is about the next session's close against today's close, not about the stock over the long run.
How is it made?
Behind it is a statistical model of the "decision trees" kind, trained on daily stock data since 2016. It needs at least 60 sessions of history before a stock enters the calculation, and it takes numbers that are all known at today's close, such as:
How is it judged?
The rule is fixed: a new close more than 0.5% above the forecast day's close is "up", more than 0.5% below is "down", and anything between is "stable". A forecast is right when its word matches what happened.
Some sessions are not judged at all: when the stock moved more than 20.5%, since that usually reflects a corporate action such as a distribution or a split, or a return from a suspension.
How do you read the record?
Beside the forecast you will find its record: how often it was right over this stock's recent forecasts, and for all stocks since day one. It is a record of results already measured, not a promise about what comes next.
See the forecast and its record
Open any stock page and look at the next-session forecast and the record beside it.
Check yourself
1. The forecast day closed at 20.00 and the new close is 19.85. The outcome?
2. How do you judge whether a hit rate is good?
Summary
- The forecast is one word for the next session: up, down or stable.
- It is judged on the close against the forecast day's close, with a 0.5% band.
- The record is measured results; read it against a simple rule.