Do Anomalies Die When They Are Published?

Abstract

McLean and Pontiff (2016) show that published stock-return anomalies weaken after the academic articles that document them appear. We revisit their question with the full Open Source Asset Pricing library—monthly long-short returns for 212 published predictors, 173,302 anomaly-months over 1926–2024—and classify each anomaly-month as in-sample, post-sample (out of sample but before publication), or post-publication using each study’s original sample-end year and journal-publication year. The average anomaly earns 0.61% per month in sample; this falls to 0.42% out of sample and to 0.30% after publication—a within-anomaly decline of-51% (paired t=-7.6). In a pooled panel with anomaly fixed effects the post-publication return is 0.27 points per month lower (t=-7.9), and the effect survives anomaly-plus- calendar-month fixed effects (-0.24, t=-3.2). The decay is thus large and robust. Four extensions sharpen the interpretation without overclaiming. First, the decay is strongly heterogeneous in observable strength: anomalies with above-median in-sample returns lose an extra-0.28 points (t=-4.3) beyond the baseline, exactly the cross-sectional pattern McLean and Pontiff (2016) predict (though partly regression to the mean). Second, because every anomaly is eventually “treated,” we build a not-yet-published-control stacked difference-in-differences: it gives a clear-0.37 (t=-5.2) with cohort-anomaly fixed effects but only-0.05 (t=-0.6) once cohort×calendar-year effects absorb the common trend—so the drop is largely an out-of-sample phenomenon, not a discrete publication kink. Third, randomization inference that permutes each anomaly’s publication date within its own out-of-sample life shows the true date is not special for the level decay (RI p = 0.31) and only marginally so for the incremental effect (p = 0.08). Fourth, a minimum-detectable-effect analysis shows the null on the risk-versus-mispricing split is informative only for large differences (≥39% of the in-sample mean), which is exactly why the pivotal hand-collection task—reading each original article’s stated economic framing—is left to the student.