Paper: arXiv 2311.14419

Authors: Deborah Miori, Constantin Petrov

Abstract

Starting from a corpus of economic articles from The Wall Street Journal, we present a novel systematic way to analyse news content that evolves over time. We leverage on state-of-the-art natural language processing techniques (i.e. GPT3.5) to extract the most important entities of each article available, and aggregate co-occurrence of entities in a related graph at the weekly level. Network analysis techniques and fuzzy community detection are tested on the proposed set of graphs, and a framework is introduced that allows systematic but interpretable detection of topics and narratives. In parallel, we propose to consider the sentiment around main entities of an article as a more accurate proxy for the overall sentiment of such piece of text, and describe a case-study to motivate this choice. Finally, we design features that characterise the type and structure of news within each week, and map them to moments of financial markets dislocations. The latter are identified as dates with unusually high volatility across asset classes, and we find quantitative evidence that they relate to instances of high entropy in the high-dimensional space of interconnected news. This result further motivates the pursued efforts to provide a novel framework for the systematic analysis of narratives within news.

Complexity vs Empirical Score

  • Math Complexity: 3.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper uses standard NLP and network analysis techniques (GPT extraction, graph theory, fuzzy clustering) with accessible formulas, lacking deep mathematical derivations. However, it demonstrates high empirical rigor with a substantial real-world dataset (Wall Street Journal, 4 years), multi-asset volatility indices, rolling z-score calculations, and a clear backtesting-ready framework for mapping news features to market dislocations.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Linking News Narratives to<br>Financial Market Dislocations"] --> B["Data Input:<br>Wall Street Journal<br>Economic Articles"]
  B --> C["Method: GPT3.5 Entity Extraction"]
  C --> D["Process: Weekly Co-occurrence<br>Graph Construction"]
  D --> E["Method: Fuzzy Community<br>Detection & Sentiment Analysis"]
  E --> F["Process: Feature Design<br>High Entropy & Topic Structure"]
  F --> G["Outcome: Quantitative Evidence<br>High News Entropy -><br>Market Volatility Dislocations"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a novel framework that uses GPT‑3.5 to extract main entities from WSJ articles, builds weekly co‑occurrence graphs, applies network analysis and fuzzy community detection to identify topics and link news structure to financial market dislocations.

Source sentence
Our research introduces a novel framework for topic identification within economic news. Indeed, we leverage on GPT3.5 to extract the main entities (both concrete or abstract) of reference for each article in an available corpus, and aggregate entities’ co‑occurrence among articles in weekly graphs.

Assumes 5

  • News articles can be meaningfully aggregated at a weekly frequency to capture evolving narratives. data
    Source sentence
    aggregate co‑occurrence of entities in a related graph at the weekly level.
  • Sentiment around main entities is a more accurate proxy for overall article sentiment. model
    Source sentence
    we propose to consider the sentiment around main entities of an article as a more accurate proxy for the overall sentiment of such piece of text
  • The week ending on 14th March 2021 is an outlier and can be dropped without affecting results. data
    Source sentence
    Importantly, we considered the week ending on 14th March 2021 as an outlier and dropped it, since we could only download three associated articles from Factiva.
  • Logistic regression is sufficient to assess the relationship between news features and market dislocations. statistical
    Source sentence
    logistic regression is indeed the final technique adopted in our work, since we simply aim to complete an initial assessment of whether news structure, especially modelled by our graph construct, can provide any enhancement to our understanding of markets.
Show 1 more
  • Sentiment is one of the most important market‑related features that can be extracted from economic news. model
    Source sentence
    Our news features limit themselves to measures of the sentiment across articles, since we can safely assume that sentiment is indeed one of the most important market‑related features that can be extracted from a plain corpus of economic news.

Data

  • Wall Street Journal economic articles · Factiva text-news commercial
    January 2020 to October 2023 · Economics section of WSJ, English language
    Source sentence
    we consider approximately four years of daily news, i.e. from January 2020 to October 2023, and aggregate them at weekly level

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Lower sentiment in news is significantly associated with weeks of market dislocation. statistical
    Source sentence
    As expected (and desired), lower sentiment within news is more likely to be associated with weeks of market dislocation.
  • Graph‑based features such as entropy, giant component ratio, eigenvector centrality and community count are significant predictors of market dislocations in both contemporaneous and predictive logistic regressions. statistical
    Source sentence
    multiple features computed from our graph construct are found to be also significant, especially from the entropy of discussions and consequent likelihood of contagion of sentiment, both in the contemporaneous and predictive scenarios.
  • The predictive logistic regression model achieves a higher pseudo R‑squared of 0.342, indicating stronger explanatory power when using current market state and news features. performance
    Source sentence
    Model: Logit Pseudo R-squared: 0.342

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 18 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 3 papers in the archive (3 by independent authors)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (30)

  1. BERTopic: Neural topic modeling with a class-based TF-IDF procedure (2022) — link
  2. Company Similarity using Large Language Models — → 2308.08031
  3. Fund2Vec: Mutual Funds Similarity Using Graph Learning (2022) — link
  4. Fast unfolding of communities in large networks (2008) — link
  5. Full-Text or Abstract? Examining Topic Coherence Scores Using Latent Dirichlet Allocation (2017) — link
  6. Uncovering fuzzy community structure in complex networks (2007)
  7. Distributed Representations of Words and Phrases and their Compositionality (2013) — link
  8. Fund2Vec: Mutual Funds Similarity using Graph Learning (2021) — link
and 22 more
  1. node2vec: Scalable Feature Learning for Networks (2016) — link
  2. Efficient Estimation of Word Representations in Vector Space (2013) — link
  3. Laplacian Eigenmaps for Dimensionality Reduction and Data Representation (2003) — link
  4. Intraday News Trading: The Reciprocal Relationships Between the Stock Market and Economic News (2018) — link
  5. How news affects the trading behaviour of different categories of investors in a financial market (2015) — link
  6. News-driven return reversals: Liquidity provision ahead of earnings announcements (2014) — link
  7. Quantifying correlation between Financial News and stocks (2016) — link
  8. High quality topic extraction from business news explains abnormal financial market volatility (2012) — link
  9. Temporal Data Meets LLM – Explainable Financial Time Series Forecasting — → 2306.11025
  10. GPT-InvestAR: Enhancing Stock Investment Strategies through Annual Report Analysis with Large Language Models — → 2309.03079
  11. Linking microblogging sentiments to stock price movement: An application of GPT-4 — → 2308.16771
  12. Top2Vec: Distributed Representations of Topics (2020) — link
  13. ChatGPT: A comprehensive review on background, applications, key challenges, bias, ethics, limitations and future scope (2023) — link
  14. Financial Market Dislocations (2014)
  15. GPT-4 Technical Report (2023) — link
  16. Modularity and community structure in networks (2006) — link
  17. SMOTE : Synthetic Minority Over-sampling Technique (2002) — link
  18. REGAL (2018) — link
  19. A Multi-Species Functional Embedding Integrating Sequence and Network Structure (2018) — link
  20. Systematic comparison of graph embedding methods in practical tasks (2021) — link
  21. VADER: A Parsimonious Rule-based Model for Sentiment Analysis of Social Media Text (2015)
  22. Realised Volatility Forecasting: Machine Learning via Financial Word Embedding (2021) — link

4 of 30 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.