Market Sentiment Analysis, with the Caveats Built In.
Classify the tone of financial news, follow how it shifts across companies, sectors and events, and drill down to the articles behind every reading.
Tone of coverage
See how the conversation around a company or sector is changing
Sentiment analysis uses natural language processing to label financial news as positive, negative, neutral or mixed, then aggregates those labels over time. Used carefully, it is a useful way to summarize large volumes of coverage. Used carelessly, it is easy to over-read.
- Article-level labels with a confidence indicator
- Trend views by company, sector and watchlist
- Event windows to study coverage before and after an announcement
- Drill-downs from any reading to the articles that produced it
Request early access
In development
In development
Product status: Quantzivo AI is in development. Features described here reflect the planned scope and may change before release.
Planned capabilities
What the sentiment module is designed to do
Financial news sentiment
Tone labels for individual news items about the companies, sectors and topics you follow.
Market sentiment classification
Positive, negative, neutral and mixed labels, each with a confidence level rather than false precision.
Sentiment trend dashboards
Rolling views of sentiment by company, sector or watchlist, shown together with the volume of coverage.
Event-based research
Compare coverage in a window before and after an event such as a results release or policy decision.
Historical sentiment analysis
Look back over longer periods to see how the tone of coverage evolved through different market conditions.
Article drill-down
Open any reading to see the individual stories, their labels and links to the original sources.
Methodology at a glance
How a sentiment reading is built
Collect and match
News items are gathered, de-duplicated and matched to the companies and sectors they discuss.
Classify
A language model labels each item's tone toward the subject it mentions and records how confident it is.
Aggregate
Labels are combined by company, sector and time window, with the number of articles shown beside each reading.
Review
You can open the underlying articles to check whether the labels make sense in context.
Classification labels
What each label means
Labels describe the tone of the text as written. They are not judgments about the company or forecasts of what will happen next.
| Label | What it means | Example phrasing (illustrative) |
|---|---|---|
| Positive | The text mainly reports favorable developments for the subject. | "Quarterly revenue rose on stronger demand and the company maintained its outlook." |
| Negative | The text mainly reports unfavorable developments for the subject. | "The company lowered its full-year guidance, citing weaker orders." |
| Neutral | The text is factual or procedural, with no clear favorable or unfavorable angle. | "The board will meet next week to consider the annual accounts." |
| Mixed | The text contains significant favorable and unfavorable points. | "Profit beat last year's level, but margins narrowed as costs increased." |
Caveats
Where sentiment analysis struggles
Every sentiment model makes mistakes. Knowing the typical failure modes helps you decide how much weight a reading deserves.
Sarcasm and irony
Commentary that says one thing and means another is often labeled by its literal words.
Ambiguity and hedging
"Results were broadly in line" can read as positive or negative depending on what was expected.
Financial language
Words like "cut", "beat" or "exposure" change meaning by context: a rate cut and job cuts are not the same story.
Thin coverage
A reading based on a handful of articles can swing sharply. Always check the article count.
Repeated stories
The same story syndicated many times can dominate an aggregate unless duplicates are grouped.
Sentiment is not direction
Positive coverage does not mean prices will rise. Markets often react to news relative to expectations.
Sentiment is not a trading signal. Sentiment readings describe the tone of published coverage. They do not predict price direction and are not buy or sell recommendations.
Example studies
Research questions sentiment can help with
Event-based
Coverage around a results release
Compare the tone and volume of coverage in the weeks before and after a company reports, then read the stories that drove any shift.
- Event window
- Article volume
- Drill-down
Historical
Sector tone through a rate cycle
Track how coverage of a rate-sensitive sector changed as central bank policy shifted, alongside the macro data for the same period.
- Sector trend
- Macro overlay
- Long history
Comparative
How peers are being covered
Put several companies in the same industry side by side to see whether one is attracting noticeably different coverage, and why.
- Peer view
- Watchlist
- Topic filters
Macro Intelligence
Backtesting
FAQs
All FAQsSentiment analysis questions
No. Sentiment describes the tone of coverage, not where prices are heading. Prices can fall on good news and rise on bad news, depending on what the market already expected.
That is the intent. Every reading is designed to drill down to the articles behind it, with their labels, confidence and links to the original source, so you can judge the classification yourself.
Researching whether sentiment adds anything to a strategy is a legitimate question, but it needs a proper backtest with realistic timing and costs. Quantzivo Studio is where that kind of testing is planned to happen.
The planned scope focuses on financial news. Coverage of other text sources has not been decided and will be confirmed closer to release.
Sentiment analysis
Measure the tone of coverage, and read what is behind it
Request early access to the sentiment module, or learn the NLP and validation methods behind it in Quantzivo Academy.