A continuity check on the forecast: does the model’s next few days point the same way, day by day, as the stock’s last few days? It once counted for 8% of the conviction score — but we backtested it and it earns nothing, so its weight is now zero. It is shown for transparency; it no longer affects the score.
Take the stock’s last five closing prices — that gives four day-to-day moves, each either up or down. Now take the first five points of the model’s freshly-produced forecast — today and the next four days — which gives four predicted moves, each also either up or down. Line the two sequences up and count how often they agree on direction.
All four agree, and path momentum is at its maximum: the forecast’s near-term shape echoes the stock’s recent rhythm. Two of four agree — no better than a coin toss — and it sits at the neutral midpoint. None agree, and it is at its minimum: the model is calling for the opposite of what the stock has just been doing, on every one of those days.
This is the part that is easiest to get wrong, so it is worth stating plainly. Both sequences are read at the same moment — the model’s forward-looking path and the stock’s recent past. Nothing here goes back to a forecast made a week ago and checks whether it came true. No such forecast enters the calculation.
So a low reading does not mean "the model was wrong last week", and a high reading does not mean "the model has been right". What it measures is agreement in rhythm: is the forecast continuing the stock’s current direction, or is it calling for a turn? If you want to know how accurate our forecasts have actually been, that question is answered with real measured numbers on the calibration page — not by this component.
Intuitively it seemed useful: when the model’s near-term path runs with the stock’s current move, two fairly independent views — the forecast and the tape — appear to agree. So it originally added a small bonus to the conviction score. The trouble is that "seems useful" is a hypothesis, not a result.
Today it is a transparency observation, not a scored signal. It still tells you something descriptive — is the forecast continuing the recent rhythm or calling for a turn — but it no longer nudges the conviction score in either direction, because we could not show that the nudge was earned. Note too what it never checked: the size of the moves. If the model’s path drifts up 0.1% while the stock rose 4%, that counted as agreement. It is a direction check, nothing more.
The 8% weight it used to carry was a guess — a design prior nobody had ever tested. So we tested it. We re-ran the model as it would have stood on 5,400 past days (the TOP-20 S&P 500 stocks, monthly through 2017–2026), recorded the path-momentum reading each time, and paired it with the return that actually followed over the next 3, 6 and 12 months.
The reading had essentially no relationship with what happened next: the rank correlation with forward return was about zero at every horizon and in every market condition, and stocks with a strong reading went on to rise no more often than those with a weak one. A component that does not predict anything should not move a money-bearing score, so we set its weight to zero and redistributed it across the components that measure up. We left the row on the page rather than quietly deleting it: being straight about what does not work is the same discipline that makes the signals that DO work trustworthy.
No — it cannot tell you that. It compares the model’s forward path with the stock’s recent moves, both read today; it never scores a past prediction against what actually happened. A weak reading means the forecast is calling for a change of direction from the stock’s recent rhythm, not that it has been proven wrong. Measured forecast accuracy lives on our calibration page.
Because the component compares direction day by day, not the total move. A stock can finish the week sharply higher while its individual daily ups and downs disagreed with the model’s predicted sequence — for instance one huge up day among three down days.
No — it cannot move the verdict at all. Its weight is zero: since we measured it and found no predictive value, it is shown for transparency but does not contribute to the conviction score. The score is driven by the seven components that do carry information, and quantum direction or quantum tail risk scoring zero still caps it at 3.0 out of 10.
Browse all S&P 500 tickers to see this metric applied to individual companies.
Educational research only — not investment advice.