Information Aggregation and P-Hacking

This paper studies the interplay between information aggregation and p-hacking in the context of predicting stock returns. The standard information-aggregation techniques exacerbate p-hacking by increasing the probability of the type I error. We propose an aggregation technique that is a simple modification of three-pass regression filter/partial least squares regression with an opposite property: the predictability tests applied to the combined predictor become more conservative in the presence of p-hacking. Using simulations, we quantify the advantages of our approach relative to the standard information-aggregation techniques. We also apply our aggregation technique to three sets of return predictors proposed in the literature and find that the forecasting ability of combined predictors in two cases cannot be explained by p-hacking.

This paper was accepted by Karl Diether, finance.