EquityBias
  • About
  • FAQ
  • Glossary
  • Plans
Sign up free
How a score is produced

Methodology

A Bias Score is a single number between -100 and +100 for one equity on one trading day. It describes where several categories of publicly available market data sit, gathered into one reading. It is a description of data that already exists. It is not a forecast, not a recommendation, and not a statement about what a price will do.

This page says what feeds that number, where the data comes from, how often it changes and where the reading breaks down. It does not publish the weighting. That is the part with commercial value, and disclosing the basis of a recommendation has never required handing over the model.

Who produces it

EquityBias is operated by Jones Inc., CVR DK34325170, Denmark. Nicky Jørgensen is responsible for the production of the readings and for the content of this page. Jones Inc. is not an investment firm, holds no licence from the Danish Financial Supervisory Authority, and is not supervised by any financial authority. It produces market analysis on its own account.

Interests and conflicts, including any position held in a specific instrument, are set out on the conflicts page, and the position in an individual name is stated on that name's own record.

What goes in

Four categories of information, each scored on its own scale before anything is combined. Each one is a different kind of evidence about the same company, which is the whole reason for using four rather than one.

Price structure. Where the price sits against its own recent history, across a short, a medium and a long window. Derived entirely from daily open, high, low, close and volume. No intraday data is used at any point.

Analyst coverage. The published ratings and price targets of the sell-side analysts covering the name, and how those have moved. A name with two analysts and a name with forty are not the same evidence, and the reading reflects that.

News sentiment. Company news headlines from the recent window, classified by a financial-language sentiment model. The classification is done per headline and aggregated; it reads tone, not truth, and it cannot tell a rumour from a filing.

Fundamentals. Reported company figures from the most recent filings: valuation, revenue and earnings trend, margins, balance sheet. These move on a quarterly cadence, so they are the slowest of the four by a wide margin.

A fifth number, the divergence score, is produced alongside the Bias Score and measures how far the four disagree with each other. A headline score near zero can mean the four agree that nothing is happening, or that two point firmly one way and two the other. Those are opposite situations and the Bias Score alone cannot tell them apart, which is what divergence is for.

Where the data comes from

  • Yahoo Finance, via the yfinance library: daily price and volume history, analyst ratings and price targets, reported fundamentals, company metadata.
  • Finnhub: company news, and price and fundamentals data where Yahoo has gaps or disagrees.
  • Yahoo Finance RSS: company news headlines for the sentiment pass.
  • FinBERT (ProsusAI/finbert), an open sentiment model trained on financial text, run by us on the headlines above. No sentiment score is bought in.

All four are third-party sources and none of them is error-free. Where they disagree about a figure, the disagreement is resolved by rule rather than by hand, and no figure is adjusted after the fact to make a score look better.

How often it changes

Scores are produced on US trading days only. Nothing is produced on a US market holiday or at a weekend, and a reading from the last trading day stands until the next one replaces it.

  • 10:00 UTC. A fast pass over roughly 158 large-cap names, used for the day's free material.
  • 12:00 UTC. The full pass over the whole universe of 550+ names, including the sentiment model. This is the run that produces the day's Bias Scores.
  • Immediately after. Subscriber dashboards refresh. That is the moment a reading is first published.
  • 3 trading days later. The same reading reaches the public pages and the twelve-month public record.

The 12:00 UTC run happens before the US open, so the newest reading always describes the previous session's close. Fundamentals inside it are as recent as the last quarterly filing, which can be nearly three months old, and that is a property of the data rather than a lag in the pipeline.

What it cannot do

It does not predict. Every input is backward-looking. A score describes the state of the evidence at a close, and there is no mechanism in it that forecasts a price, a direction or a date.

It is blind to anything not yet in the data. An announcement made after the close, a filing published this morning, a rumour that has not reached a headline: none of it is in the reading until the next run picks it up.

Thin coverage produces thin readings. A small company with two analysts and little news has two of the four categories running on almost no evidence. The score still comes out on the same scale, which makes it look as solid as a reading on a large-cap name. It is not.

The sentiment pass reads tone. It classifies the language of a headline. It does not know whether the headline is accurate, whether it is about a lawsuit already priced in, or whether the same story is being counted from three outlets.

ETFs are not companies. The universe includes around twenty ETFs, where fundamentals and analyst coverage mean something different from what they mean for a single company. Read those with that in mind.

Source errors pass through. A wrong price target or a mis-tagged filing at the source becomes a wrong input here. Checks catch the obvious cases; they do not catch everything.

Past readings say nothing about future results. The twelve-month record for each name is published so that claim can be checked rather than taken on trust.

The record

Every reading published on an instrument in the last twelve months is available on that instrument's own record page, outside the paywall, with the date it was produced, the date it was first published, the closing price at the time and the reading that superseded it. From any ticker page, follow the link to its score history.

What is deliberately not here

The weighting between the four categories, the thresholds, the windows, and the way the divergence score is derived. Those are the product. The disclosure obligation is that a reader can tell what kind of evidence a reading rests on and how far to trust it, and that is what the sections above are for.

Risk warning

EquityBias is a market analysis tool operated by Jones Inc. (CVR DK34325170). A Bias Score is a description of data the market has already produced, gathered into one number. It is not investment advice, not a personal recommendation, and not a forecast of price.

Trading equities carries risk and you can lose part or all of the money you put in. Past readings say nothing about future results. Decide for yourself, and take professional advice if you need it.

Last updated: September 2026

EquityBias

Market Bias, Quantified. EquityBias produces daily Bias Scores via a proprietary multi-layer scoring engine, not financial advice.

$
Your stocks, scoredGet a free Bias Score report on up to five names you own or follow, then keep watching them on your free dashboard →
Explore
  • Free watchlist report
  • Stocks
  • Insights
  • Glossary
Company
  • About
  • FAQ
  • Plans
  • Sign in
Contact

Operated by Jones Inc.
CVR DK34325170
equitybias.com

  • contact@equitybias.com
Legal
  • Privacy Policy
  • Cookies
  • Terms of Use
  • Disclaimer
  • Methodology
  • Interests and conflicts

© 2026 EquityBias · Operated by Jones Inc. (DK34325170)

equitybias.com

EquityBias is a market analysis and research tool. Scores represent aggregated market data and do not constitute financial advice, investment recommendations, or price predictions.