NOSIBLE Launches Semantic Factors for Stock Risk Research

In Short

NOSIBLE has rolled out Semantic Factors, a new way to convert web text into daily “semantic factors” and stock-specific betas for market research.

What Happened

NOSIBLE introduced Semantic Factors, a feature set and open source toolkit that turns large-scale web text into daily risk indicators you can inspect and export.

A semantic factor is a time series, meaning a daily score that tracks how strongly a concept shows up in web text. Users define the concept in plain sentences, set a polarity, meaning whether the factor should rise on “more of” something or “less of” it, then compute scores.

NOSIBLE’s workflow has three steps. First, define “relevance anchors” and “polarity pairs,” which are short phrases that describe what the model should look for. Second, score those definitions using NOSIBLE World and an external model routing layer via OpenRouter. Third, export charts, code, and CSV files for analysis.

The company also explains how to convert a semantic signal into stock betas. A beta here is a statistical sensitivity estimate, meaning how a stock tended to move when the factor changed, after controlling for market returns. NOSIBLE highlights that this is association, not proof of cause.

On its Semantic Factors page, NOSIBLE lists 30 prebuilt factors, including geopolitical risk, sanctions and export controls, inflation attention, recession nowcast, and supply-chain pressure.

Why It Matters

For African investors, analysts, and fintech teams, text-based risk signals can be a practical input for research. Many market-moving events show up in news and online commentary before they show up in traditional datasets.

NOSIBLE’s pitch is transparency and repeatability. By letting users define concepts in normal language and export the underlying time series, teams can audit the signal and rerun it when models or assumptions change.

It also lowers the barrier for internal research. Instead of building custom natural language processing pipelines, teams can start from a packaged set of factors and focus on testing how those signals relate to portfolios, sectors, or specific listings.