Research Reports
Every Sunday, a PDF with the week’s news picture: read by AI, checked against market prices.
Methodology
Data sources: Financial news from more than 90 sources in four languages, collected from their public RSS feeds.
AI analysis: Each article is read by a large language model — by day Qwen 3.8 27B on our own hardware, overnight OpenAI GPT-6 Luna. It assigns a tone from −1.0 to +1.0, the market the article is about, a relevance score and sector tags.
Checking: We compare each headline’s tone with the actual price move one to seven days later, using Yahoo Finance prices, and measure how often the two matched.
In the report: Average tone per market with 95% confidence intervals, correlations between sectors and how often the news matched the later price move.
How to Cite These Reports
Use the following citation format to reference our weekly PDF reports in your research papers, articles, or presentations. The citation includes the specific report date and direct PDF download URL.
APA Style (7th Edition)
BibTeX Format
Fair Use Statement
These reports contain historical analysis of publicly available news feeds. Our AI-extracted sentiment scores, statistical summaries, and evaluations constitute transformative fair use under § 87f UrhG (German Copyright Act) and US Fair Use doctrine (17 U.S.C. § 107) for purposes of research, criticism, and education. Original news articles are linked and attributed to their respective publishers.
Available Publications
Loading reports...
For Researchers
Data Access: Researchers requiring access to raw data, model outputs, or custom analysis periods can contact us at [email protected].
API Access: Academic institutions can request API access for automated data retrieval. Rate limits and quotas will be negotiated based on research scope.
Collaboration: We welcome collaborations with universities and research institutions. Please include your institutional affiliation and research proposal in your inquiry.
License: CC BY-NC-SA 4.0 (Attribution-NonCommercial-ShareAlike)