Does Academic Research Destroy Stock Return Predictability?
McLean and Pontiff show that published anomaly results weaken outside original samples and after publication, requiring horizon and replication caveats around any public pattern.
McLean and Pontiff study 97 variables that prior academic work linked with cross-sectional stock returns. They compare each relation inside the original research sample, after that sample ends, and after the result becomes public. Portfolio returns are lower out of sample and lower again after publication in their analysis. The authors use the two declines to separate an upper-bound estimate of data-mining effects from changes associated with publication-informed trading.
The paper is not specific to Form 4 or Form 13F, but it is an essential methodological companion. A historical association can weaken when the sample period changes, when investors learn about it, when trading costs are incorporated, or when the original specification was selected from many alternatives. Publicizing an anomaly can also change trading volume and correlations among portfolios tied to published characteristics. A static literature result is therefore not a timeless company-level rule.
Insider Atlas uses old and new studies to explain public filings, not to claim that a published result remains implementable at its original magnitude. The site does not present a backtest, company-level outcome, or strategy score. McLean and Pontiff also support a lower evidence grade for raw institutional crowding: once a characteristic is widely known, 13F co-movement may document attention and common positioning as much as independent information. Any later return study would need a true out-of-sample design, release-date information set, delisting-aware universe, and explicit costs.
Abstract excerpt
“We study the out-of-sample and post-publication return predictability of 97 variables shown to predict cross-sectional stock returns.”
Connections in Insider Atlas
The NTSK cluster and NTSK profile are current observations, not an out-of-sample validation of any cited historical magnitude.