Paper: arXiv 2507.04481

Authors: Paul Glasserman, Kriste Krstovski, Paul Laliberte, Harry Mamaysky

Abstract

Over the past 30 years, nearly all the gains in the U.S. stock market have been earned overnight, while average intraday returns have been negative or flat. We find that a large part of this effect can be explained through features of intraday and overnight news. Our analysis uses a collection of 2.4 million news articles. We apply a novel technique for supervised topic analysis that selects news topics based on their ability to explain contemporaneous market returns. We find that time variation in the prevalence of news topics and differences in the responses to news topics both contribute to the difference in intraday and overnight returns. In out-of-sample tests, our approach forecasts which stocks will do particularly well overnight and particularly poorly intraday. Our approach also helps explain patterns of continuation and reversal in intraday and overnight returns. We contrast the effect of news with other mechanisms proposed in the literature to explain overnight returns.

Complexity vs Empirical Score

  • Math Complexity: 4.5/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 8.0)

Why this score: The paper employs advanced NLP and statistical methods (topic modeling, supervised selection, OLS regressions with clustering) but focuses heavily on empirical backtesting with 2.4M articles and out-of-sample forecasts, making it data- and implementation-heavy rather than mathematically dense.

Research Flowchart

  flowchart TD
  A["Research Goal: <br>Explain the 'overnight premium' <br>(why stock gains occur overnight)"] --> B["Data Collection & Processing"]
  B --> C["Novel Supervised Topic Analysis"]
  C --> D["Identify Market-Explanatory Topics"]
  
  B -->|"Input:"| B1["2.4M News Articles"]
  B -->|"Input:"| B2["Stock Price Data (Overnight/Intraday)"]
  
  C -->|"Method:"| C1["Select topics maximizing <br>return explanation"]
  C -->|"Method:"| C2["Measure topic prevalence & <br>market response"]
  
  D --> E["Out-of-Sample Forecasting"]
  E --> F["Key Findings/Outcomes"]
  
  F --> F1["News explains significant <br>portion of overnight premium"]
  F --> F2["Intraday vs. Overnight <br>News patterns differ"]
  F --> F3["Topics predict specific <br>stock overnight performance"]

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 supervised topic analysis method that selects news topics based on their ability to explain contemporaneous market returns and uses it to forecast stock returns.

Source sentence
We apply a novel technique for supervised topic analysis that selects news topics based on their ability to explain contemporaneous market returns.

Assumes 5

  • Dividends are included in close-to-open returns. data
    Source sentence
    Dividends are paid to the owner of the shares at the close, so we include them in close-to-open returns.
  • All stock splits and buybacks take effect at the close. data
    Source sentence
    We also assume that all stock splits and buybacks take effect at the close.
  • Firm-level news exposures are persistent over time. model
    Source sentence
    Firm-level news exposures are persistent, which allows us to use a stock’s past news exposure, combined with the market’s past response to that news exposure, to estimate how that stock will do in future intraday and overnight periods.
  • Articles mentioning more than three S&P 500 companies are dropped because they discuss overall market developments rather than company‑specific news. data
    Source sentence
    drop articles that mention more than three S&P 500 companies because these usually discuss overall market developments rather than company-specific news.
Show 1 more
  • The intraday and overnight returns are calculated from open‑to‑close and close‑to‑next‑open respectively, with no adjustment for trading frictions. execution
    Source sentence
    We calculate intraday returns from the open to the close each day; we calculate overnight returns from the close to the next open.

Data

  • S&P 500 firm price data · CRSP market-prices commercial
    1996‑2022 · S&P 500 constituents
    Source sentence
    We obtain daily open and close stock prices from CRSP.
  • Thomson Reuters News Feed Direct archive · Thomson Reuters text-news commercial
    1996‑2022 · Articles mentioning S&P 500 companies
    Source sentence
    Our news text comes from the Thomson Reuters News Feed Direct archive (hereafter TR).

Recognised sources: CRSP

Costs

Not modeled A trading or execution result is reported without costs.

Source sentence
Because of the extreme turnover required to trade the over-intra effect, our findings fall short of being a viable trading strategy.

Finds 5

  • A large part of the overnight‑over‑intraday return gap can be explained by features of intraday and overnight news. statistical
    Source sentence
    We find that a large part of this effect can be explained through features of intraday and overnight news.
  • Both time variation in news‑topic prevalence and differences in market responses to those topics contribute to the intraday‑overnight return difference. statistical
    Source sentence
    We find that time variation in the prevalence of news topics and differences in the responses to news topics both contribute to the difference in intraday and overnight returns.
  • The model forecasts which stocks will do particularly well overnight and particularly poorly intraday in out‑of‑sample tests. predictive-accuracy
    Source sentence
    In out-of-sample tests, our approach forecasts which stocks will do particularly well overnight and particularly poorly intraday.
  • News flow partly explains the over‑intra correlations documented in Table 1. statistical
    Source sentence
    news flow partly explains the over-intra correlations in Table 1.
  • Intraday‑selected stocks have significantly negative returns, confirming the predicted underperformance. performance
    Source sentence
    The estimates of bSS show that the intraday selected stocks have significantly negative returns, as predicted.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 34 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive (1 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 (40)

  1. Overnight returns and firm-specific investor sentiment (2018)
  2. Overnight returns, daytime reversals, and future stock returns (2022)
  3. Illiquidity and stock returns: Cross-section and time-series effects (2002)
  4. Price discovery and trading after hours (2003)
  5. What drives momentum and reversal? Evidence from day and night signals (2022)
  6. The night and day of A mihud’s (2002) liquidity measure (2021)
  7. Paying attention: overnight returns and the hidden cost of buying at the open (2012)
  8. A Lottery-Demand-Based Explanation of the Beta Anomaly (2017)
and 32 more
  1. Alternative Transformations to Handle Extreme Values of the Dependent Variable (1988)
  2. Latent dirichlet allocation (2003)
  3. The cross-section of intraday and overnight returns (2021)
  4. Market return around the clock: A puzzle (2023)
  5. Information, trading, and volatility: Evidence from firm-specific news (2019)
  6. The Overnight Drift (2023)
  7. lda: Collapsed Gibbs sampling methods for topic models (2015)
  8. Return differences between trading and non-trading hours: Like night and day (2008)
  9. Choosing news topics to explain stock market returns (2020)
  10. Time Variation in the News–Returns Relationship (2025)
  11. The characteristics that provide independent information about average US monthly stock returns (2017)
  12. Liquidity and market structure (1988)
  13. Asset pricing: A tale of night and day (2020)
  14. Returns in trading versus non-trading hours: The difference is day and night (2011)
  15. Narrative Economics (2019)
  16. Narrative asset pricing: Interpretable systematic risk factors from news text (2023)
  17. Business cycle narratives (2019)
  18. A tug of war: Overnight versus intraday expected returns (2019)
  19. Heterogeneous liquidity providers and night-minus-day return predictability (2023)
  20. Night Moves: Is the Overnight Drift the Grandmother of all Market Anomalies? (2024)
  21. Probabilistic topic models (2007)
  22. Supervised topic models (2007)
  23. Reading tea leaves: how humans interpret topic models (2009)
  24. Dynamic topic models (2006)
  25. Choosing News Topics to Explain Stock Market Returns (2020)
  26. Online polylingual topic models for fast document translation detection (2013)
  27. Online multilingual topic models with multi-level hyperpriors (2016)
  28. Bart-tl: Weakly-supervised topic label generation (2021)
  29. Finding scientific topics (2004)
  30. Probabilistic topic models (2007)
  31. Chat-GPT (2023) — link
  32. GPT-4 Technical Report (2023) — link

0 of 40 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.