How the ChipSentiment score is calculated
6 August 2026 · 6 min read
Every stock on the board carries a number between 2 and 98. This page explains exactly how it is produced, because a sentiment score you cannot audit is a sentiment score you should not trust.
What the score is, and what it isn't
It is a description of how a stock is currently behaving relative to its own history and to its peer group. It is not a valuation, not a forecast, and not a recommendation. A score of 90 does not mean a stock is a good buy — it frequently means the opposite, since it takes a large recent move to get there.
The six components
Each component contributes points to a base of 50. The maximum contribution of each is fixed and published.
1-day momentum, ±18
Today's percentage change, scaled. The largest single input, because a sentiment read that ignores what just happened is not describing sentiment.
5-day trend, ±13
The five-session drift. Separates a one-day spike from a sustained move.
52-week position, ±11
Where the current price sits between the 52-week low and high. A stock at the top of its range is in a different psychological position from one near its low, even if both moved 2% today.
Participation, ±8
Today's volume against the stock's own 20-day average. A 3% move on double normal volume means something different from a 3% move on a quiet day.
Strength versus the complex, ±10
The stock's five-day move minus the market-cap-weighted five-day move of all 34 names on the board.
This is the component I would defend hardest. In a week where the entire AI silicon complex is up 8%, being up 8% tells you nothing at all. Relative strength is the only input that distinguishes a stock that is leading from one that is simply floating on the tide, and it is the reason a sector board is more useful than 34 separate quote pages.
Coverage intensity, ±10
How many articles were published about the company in the last 24 hours, measured against the median of its own previous 30 days, then signed by the direction of the five-day trend.
Note what this deliberately does not do: it does not score headlines for positive or negative tone. Most financial headlines describe the price move — "Micron slides 4% on pricing fears" is a restatement of the chart. Scoring that for polarity would feed price back into the model disguised as independent information, so the component would look orthogonal while actually being momentum counted twice.
Coverage volume genuinely is independent of price. A stock up 2% with normal attention is in a different situation from one up 2% with five times its usual article count. The direction comes from price; the news only tells us how much the market is watching.
Why the total is compressed rather than clipped
The six components are summed and then passed through a compression curve rather than simply capped at 98.
An earlier version clipped the total, and the result was that five different stocks all scored exactly 98 — the ceiling destroyed all resolution at precisely the end of the scale readers look at first. Compression means extremes approach the boundaries without landing on them, so the hottest names stay distinguishable from each other.
What happens when data is missing
Any component whose input is unavailable contributes zero rather than falling back to an older value. If a data provider stops supplying five-day history, the trend component drops out and the affected figures show a dash.
This is deliberate. Displaying last week's number in today's column, styled identically to live data, is worse than displaying nothing — the reader has no way to know which they are looking at.
Known limitations
- It is momentum-weighted. Four of the six components derive from price. On a sharp reversal the score will lag.
- Coverage intensity is thin for small names. A stock with two articles a month can look like a surge on three.
- It has no fundamental input. No earnings, margins or valuation. A company can be scored 90 while getting materially worse.
- It is not backtested yet. We have not published evidence that the score predicts forward returns, and until we do you should assume it does not.
Changes to the formula
When the formula changes, this page changes with it, and the change is noted. If you ever see a score you cannot reconcile with this page, that is a bug worth reporting.