Summary

This paper by Ran Tao, Chris Brooks, and Adrian Bell, published in 2020, explores how news sentiment is assimilated into stock prices over time, highlighting a significant divergence in how different stocks respond to news. Using unique firm-level news data spanning from 1979 to 2016, the authors distinguish between Slowly Incorporated (SI) and Quickly Incorporated (QI) news, examining their respective impacts on stock returns.

The study finds that stocks associated with SI news exhibit higher future returns than those aligned with QI news, yielding a monthly abnormal return of 139 basis points. These returns remain statistically significant after accounting for conventional risk factors and estimated trading costs. The research reveals that SI news often corresponds to low-attention stocks, providing a potential edge for investors.

Key Ideas

Page 9, Paragraph 1

The authors describe a methodology using a rolling regression approach over a 36-month window to assess how well news sentiment scores are incorporated into contemporaneous stock returns. If a stock’s price does not significantly react to news sentiment, the information is deemed slowly incorporated (SI). This classification helps identify when the market might underreact to new information, providing a basis for profitable trades.

Page 11, Paragraph 3

This paragraph presents strong empirical support for a trading strategy that capitalizes on the SI vs. QI news classification. A long-short portfolio—long on SI news stocks and short on QI news stocks—yields persistent and significant abnormal returns, even after adjusting for multiple risk models (e.g., Fama-French factors), and remains profitable after accounting for trading costs. This finding underlines the economic viability of using news-based anomalies for return predictability.

Data and Fields Used

  • Source: Dow Jones Newswire Archive (1979–2016)
  • Stock Universe: Common stocks listed on NYSE, AMEX, and NASDAQ
  • News Fields: Sentiment scores (based on Loughran and McDonald dictionary), article length, textual complexity
  • Stock Data Fields: Monthly returns, firm size, trading volume, media coverage, analyst coverage, Google Search Volume Index, Bloomberg AIA
  • Control Variables: Size, Book-to-Market ratio, Momentum, Beta, Idiosyncratic Volatility, Illiquidity

Potential Issues and Supporting Details

  1. News Sentiment Interpretation Reliability

    • Sentiment scores depend on dictionary-based methods which may not capture context-specific meanings.
    • Ambiguous or domain-specific words can be misclassified, reducing the accuracy of sentiment scoring.

  2. Temporal Alignment of News and Returns

    • The monthly aggregation may dilute the real-time impact of breaking news.
    • Intraday or daily frequency data could yield more precise insights but are not considered.

  3. Market Reaction Assumptions

    • Assumes homogenous investor behavior towards news, which overlooks variation in investor sophistication.
    • Limited attention might not uniformly apply across all market conditions or firm types.

  4. Data Skewness and Coverage Bias

    • News volume is unevenly distributed across firms, skewing results toward large-cap stocks.
    • Early years of data (pre-1995) have less news coverage, potentially impacting robustness across time.

Reflection

This research offers a novel perspective on information dissemination in financial markets. By identifying when markets underreact to news, the paper provides a practical framework for constructing profitable trading strategies. It bridges insights from behavioral finance—like limited attention—and empirical asset pricing, suggesting that market inefficiencies persist even in the face of accessible information. The robustness of findings across time periods, sentiment measures, and transaction cost estimates underscores the practical implications for institutional and algorithmic traders.

Reference