Does Peer-Reviewed Research Help Predict Stock Returns?

Paper: arXiv 2212.10317

Authors: Andrew Y. Chen, Alejandro Lopez-Lira, Tom Zimmermann

Abstract

Mining 29,000 accounting ratios for t-statistics $> 2.0$ leads to cross-sectional return predictability similar to the peer review process. For both, $\approx50%$ of predictability remains after the original sample periods. This finding holds for many categories of research, including research with risk or equilibrium foundations. Only research agnostic about the theoretical explanation for predictability shows signs of outperformance. Our results imply that inferences about post-sample performance depend little on whether the predictor is peer-reviewed or data mined. They also have implications for the importance of empirical vs theoretical evidence, investors’ learning from academic research, and the effectiveness of data mining.

Complexity vs Empirical Score

• Math Complexity: 3.5/10
• Empirical Rigor: 6.0/10 (capped from 8.5: transaction costs not modelled; see penalty flags)
• Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: This paper presents a strong empirical critique of academic factor discovery by contrasting peer-reviewed predictors against automated data mining across 29,000 accounting ratios. While the mathematical framework relies on standard statistical testing rather than complex derivations, the empirical methodology and insightful conclusions on post-sample decay offer exceptional value to quantitative finance.

Research Flowchart

  flowchart TD
    A[Research Goal: Does peer review predict stock returns better than data mining?] --> B[Data Inputs: 29,000 Accounting Ratios & Peer-Reviewed Anomaly Literature]
    B --> C[Methodology: Screen Ratios for t-stat > 2.0 vs. Academic Predictors]
    C --> D[Evaluate Post-Sample Cross-Sectional Return Predictability]
    D --> E[Finding 1: ~50% Predictability Decay Post-Sample for BOTH Groups]
    D --> F[Finding 2: Only Theory-Agnostic Research Shows Minor Outperformance]
    E & F --> G[Outcome: Peer Review Performs Similarly to Data Mining]