semantic-factors turns NOSIBLE World text into daily semantic factors. Define a concept in sentences, choose polarity, and export the factor for research.
Recompute access: NOSIBLE API key + OpenRouter API key
What semantic-factors does?
Define
Write relevance anchors and polarity pairs in plain text.
Score
Use NOSIBLE World and OpenRouter to score the definitions.
Export
Inspect the curves, code, and CSV in your own workflow.
STEP 1: WEB TEXT TO RISK
Use sentences to turn web text into a daily semantic factor you can inspect, export, and rerun.
War, terror, and conflict as a share of world coverage.
STEP 2: SIGNAL TO BETAS
Use aligned returns to turn that signal into stock betas you can compare, explain, and use directly.
Choose one semantic factor and keep its scope, threshold, and history fixed across one clearly defined measurement window.
Put the factor and adjusted closing prices on the same weekly clock, then calculate matching weekly returns and factor changes.
Regress the stock return on the market return and the change in the semantic factor over the same weekly window for each company.
A beta is a conditional historical association, not a claim that the factor caused the return or that the relationship will persist.
Estimated weekly response to a one-standard-deviation GPR rise after market control.
The regression
ri,t = αi + βmarket,i rmarket,t + βsemantic,i ΔFt + εi,tUse the factor change when the question is sensitivity to a new shock. Standardize that change so the semantic beta reads as the stock return response to one typical factor move. Add sector, country, or other controls when the research question requires them.
What the sign means
The stock tended to rise when the factor rose, after the market control.
The sample does not show a strong incremental relationship to this factor.
The stock tended to fall when the factor rose, after the market control.
Ten-stock example
Semantic beta is the stock return response to a one-standard-deviation weekly change in the 30-day log1p geometric mean of the 0.25-0.50 k=6 tanh-squashed daily semantic factor after controlling for the weekly SPY.US return. The chart adds the 7-day arithmetic average of the same 30-day geometric mean for presentation. The values are full-sample OLS estimates from EODHD adjusted_close adjusted closes, not a backtest or recommendation.
Build your next semantic factor
Get World access to recompute the examples, download all 30 semantic factors, or read the implementation on GitHub.