Non-Standard Errors in Asset Pricing: Mind Your Sorts

Abstract

Non-standard errors capture uncertainty due to differences in research design choices. We establish considerable variation in the design choices made by researchers when constructing asset pricing factors. By purposely data mining over two thousand different versions of each factor, we find that Sharpe ratios vary substantially within a factor due to different construction choices, which results in sizable non-standard errors. We provide simple suggestions that reduce the average non-standard error by 70%. Our study has important implications for model selection exercises.

Keywords: portfolio construction, factor investing, equity factors, asset pricing models, non-standard errors, p-hacking, data-mining.

JEL Classification: G11, G12, G15