We use sophisticated arbitrageurs’ observed trading as an informed indicator of mispricing to classify the 212 factors of the factor zoo, conservatively identifying 29 as mispricing. A Common Mispricing Factor built from a high-conviction core of these factors is unspanned by every leading model, and a three-factor model (MKT, SMB, CMF) prices the unclassified factors as well as leading benchmark models.
I develop a consumption-based asset pricing model in which investors are loss averse and form diagnostic expectations. With empirically plausible degrees of loss aversion and diagnosticity, the model jointly accounts for the equity premium, excess volatility, volatility clustering, and the asymmetric volatility phenomenon.
I infer the states that govern time variation in expected returns directly from 212 anomaly portfolios, using an autoencoder and a hidden Markov model. Anomaly premia exhibit strong common state dependence, and the learned states predict subsequent factor performance.