Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Measuring economic outlook in the news

Elliot Beck, Franziska Eckert, Linus Kühne, Helge Liebert, Rina Rosenblatt-Wisch

TL;DR

NEOS develops a resource-efficient, on-premises sentiment indicator for Switzerland by combining document embeddings with synthetic training data generated by large language models. It processes about $27$ million news articles and trains a logistic model with $P(y=1|x)=\sigma(x^\top w)$ on embeddings to yield monthly NEOS values, which are aggregated across articles. NEOS outperforms survey-based benchmarks and dictionary methods in GDP forecasting, yields timelier signals during crises, and provides interpretable topic-level drivers. The approach is modular, privacy-preserving, and adaptable to other data sources or jurisdictions.

Abstract

We develop a resource-efficient methodology for measuring economic outlook in news text that combines document embeddings with synthetic training data generated by large language models. Applied to 27 million news articles, the resulting indicator significantly improves GDP growth forecast accuracy and captures sentiment shifts weeks before official releases, proving particularly valuable during crises. The indicator outperforms both survey-based benchmarks and traditional dictionary methods and is interpretable, allowing identification of specific drivers of economic sentiment. Our approach addresses key institutional constraints: it performs sentiment classification locally, enabling analyses of proprietary news content without transmission to external services while requiring minimal computational resources compared to direct large language model classification.

Measuring economic outlook in the news

TL;DR

NEOS develops a resource-efficient, on-premises sentiment indicator for Switzerland by combining document embeddings with synthetic training data generated by large language models. It processes about million news articles and trains a logistic model with on embeddings to yield monthly NEOS values, which are aggregated across articles. NEOS outperforms survey-based benchmarks and dictionary methods in GDP forecasting, yields timelier signals during crises, and provides interpretable topic-level drivers. The approach is modular, privacy-preserving, and adaptable to other data sources or jurisdictions.

Abstract

We develop a resource-efficient methodology for measuring economic outlook in news text that combines document embeddings with synthetic training data generated by large language models. Applied to 27 million news articles, the resulting indicator significantly improves GDP growth forecast accuracy and captures sentiment shifts weeks before official releases, proving particularly valuable during crises. The indicator outperforms both survey-based benchmarks and traditional dictionary methods and is interpretable, allowing identification of specific drivers of economic sentiment. Our approach addresses key institutional constraints: it performs sentiment classification locally, enabling analyses of proprietary news content without transmission to external services while requiring minimal computational resources compared to direct large language model classification.

Paper Structure

This paper contains 30 sections, 9 equations, 15 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Methodology
  4. Generate document embeddings
  5. Select relevant articles
  6. Generate synthetic articles
  7. Train the sentiment model
  8. Measure economic outlook
  9. Interpretability of NEOS
  10. Keyword-based
  11. Classification-based
  12. Clustering-based
  13. Results
  14. Forecast evaluation
  15. Importance of timeliness
  16. ...and 15 more sections

Figures (15)

  • Figure 1: Number of all articles over time, broken down by publication type (print vs. online).
  • Figure 2: Overview of the computation steps for NEOS.
  • Figure 3: Number of relevant articles over time, broken down by publication type (print vs. online).
  • Figure 4: UMAP visualization of the synthetic articles with positive and negative economic outlooks.
  • Figure 5: NEOS and benchmark indicators together with Swiss real GDP growth (year-on-year, adjusted for sports events; right-hand scale). All indices are standardized.
  • ...and 10 more figures