Significance thresholds play an important role in the *discovery* of new asset pricing factors. In their papers on the cross-section of expected returns, Harvey et al. make the case for higher bars, e.g., t-statistics above 3.0 (in absolute value) instead of 2.0. It has become customary to rely on multiple testing adjustments to account for too many positive results. Often, these techniques are based on bootstrapping: they re-sample returns for instance. In the new version of my paper "Forking paths in empirical studies" (https://lnkd.in/g5f-FhYM), I advocate to use many different ways to define factors in order to generate more diverse sets of returns. This considerably increases the significance thresholds…Similar results were obtained by Soebhag et al. in their recent paper: https://lnkd.in/e-E3NptdInteresting times in the asset pricing factor space...