This paper empirically tests whether peer-reviewed economic theory helps identify cross-sectional stock return predictors that are robust out-of-sample. The authors classify 199 published predictors into three groups based on their theoretical explanation (risk, mispricing, or agnostic) and compare their post-sample performance to a set of naively data-mined predictors constructed from accounting variables. The main finding is that all groups exhibit approximately 50% post-sample decay, with no significant advantage for theory-based predictors over data-mined ones. This result holds even when focusing on risk-based predictors, which are often thought to be more stable. The authors also find that only 18% of published predictors are attributed to risk, while the majority are linked to mispricing or are agnostic. The event-time patterns of returns are strikingly similar between published and data-mined predictors, suggesting that academic research may be largely documenting patterns that emerge from data mining rather than providing unique insights. The paper concludes that peer review does not effectively protect against data mining bias, and that naive data mining can produce predictors with comparable out-of-sample performance to those published in top journals.