McLean/Pontiff (2016) give evidence that portfolio returns are 26% lower out-of-sample and 58% lower after publishing an academic paper on variables, which are likely to predict cross-sectional returns:
Our findings suggest that investors learn about mispricing from academic publications. [...] This result is also consistent with the idea that academic research draws trading attention to the predictors.
Isn't this a self-fulfilling prophecy, having regard to the data-snooping bias?
Assume data-snooping leads us to a variable which explains the cross-section of stock returns and luckily passes some robustness tests. After publishing our paper, the market adopts our findings and diminishes strategies based on our variable - although there is no economic reason to do so! In fact, anyone thinks of a previous mispricing but only after our publication real mispricing arises.