Abstract
Hundreds of cross-sectional return predictors populate the factor zoo, yet whether each reflects compensation for systematic risk or mispricing remains an open question. We classify the factor zoo using sophisticated arbitrageurs’ observed trading as an informed indicator of mispricing, and apply it to the 212 factors of the Chen and Zimmermann (2022) open-source asset pricing database, conservatively classifying 29 as mispricing. Our classification recovers the building blocks of the Stambaugh–Yuan (SY) and Daniel–Hirshleifer–Sun (DHS) mispricing factors without using any information from either model. A Common Mispricing Factor (CMF) built from a high-conviction core of nine mispricing factors earns 0.77% per month, is unspanned by every leading model, and subsumes the HML, RMW, CMA, and MOM premia. A parsimonious three-factor model (MKT, SMB, CMF) prices the 183 unclassified factors as well as leading benchmark models (FF5+MOM, q5, SY, DHS) and leaves the fewest factors unpriced. The results hold out of sample over 2016–2024 and are robust to alternative classification rules and model specifications.