Summary

This paper, The Momentum of News by Ying Wang, Bohui Zhang, and Xiaoneng Zhu (2018), investigates a novel phenomenon termed “news momentum”—the tendency of firm-specific news sentiment to persist over time. By constructing monthly firm-level sentiment scores from a large dataset of real-time news articles (2000–2016), the authors demonstrate that firms with more positive (or negative) news tend to generate similar sentiment news in subsequent periods.

The research evaluates three hypotheses to explain this persistence:

  • Stale Information Hypothesis
  • Strategic Disclosure Hypothesis
  • Fundamental Persistence Hypothesis

The results strongly support the Fundamental Persistence Hypothesis: news sentiment mirrors enduring firm fundamentals rather than merely being a product of repeated stories or manipulated disclosures.

A trading strategy exploiting this momentum—going long on firms with positive sentiment and short on those with negative—produces an annualized risk-adjusted return of 7.45%, indicating mispricing due to investor underreaction.

Key Ideas

Page 1, Paragraph 3
The paper sets the stage by emphasizing a key question: is news itself predictable in a cross-sectional sense? If so, understanding news dynamics could deepen insights into stock return anomalies and investor behavior.

Page 19, Paragraph 2
Strategic disclosure is evaluated as a potential driver of news momentum. However, empirical results show that firms’ disclosure behaviors, even when strategically managed, do not significantly explain the persistent pattern in news sentiment.

Page 44, Paragraph 1
The authors apply a trading strategy based on news sentiment scores and find that it yields strong returns not explainable by standard risk models. This reinforces the notion of market inefficiency due to investor underreaction to news patterns.

Data and Methodology

  • Source: RavenPack News Analytics (2000–2016)
  • Sample: ~2,600 firms/month across NYSE, Amex, Nasdaq
  • Articles: Over 6.7 million firm-specific news items
  • Fields Used:
    • News sentiment scores (scaled -1 to 1)
    • Firm fundamentals: ROA, earnings surprises (SUE)
    • Market data: returns, beta, size, analyst coverage, institutional holdings
    • Categorization into hard vs. soft news

Potential Issues and In-depth Analysis

  1. Reliability of Sentiment Scores

    • Opaque Algorithms: RavenPack’s scoring process is proprietary, making it a ‘black box’ and hard to independently validate.
    • Replication Challenge: Although a simpler sentiment score is tested, the original results rely heavily on a non-transparent system.

  2. Attribution to Fundamentals

    • Causality Ambiguity: The link between sentiment and fundamentals could be endogenous—better fundamentals attract positive media, rather than sentiment driving performance.
    • Measure Proxy Limitations: Use of ROA and earnings surprise may not fully encapsulate the complex drivers behind fundamental performance.

  3. Market Reaction Interpretation

    • Behavioral Biases Assumption: The analysis attributes underreaction to investors without direct behavioral evidence.
    • Alternative Explanations: Factors like transaction costs, risk constraints, and institutional trading behavior could also contribute to observed return patterns.

  4. Generalizability of Results

    • Market-Specific Bias: Data is U.S.-centric and may not hold across different regulatory or informational environments.
    • Time Frame Dependence: The 2000–2016 period includes unique market cycles (dot-com bust, 2008 crisis) possibly impacting general applicability.

Reflection

This study presents compelling evidence that market participants do not fully incorporate information from news patterns into prices. It bridges behavioral finance and information theory by showing that predictable patterns in news—rooted in real firm fundamentals—are not efficiently priced, providing both a trading edge and a deeper understanding of financial market anomalies.

Reference