Circus Ring to Zoo to Museum: The Fragility of Factors in Characteristic-based Asset Pricing Models

Abstract

Economically relevant factors in asset pricing models should impound information on the future path of state variables that drive asset risk premia. Imposing this condition, we investigate which publicly available characteristics predict individual stock returns during the sample period used by Fama and French (1993) i.e. their discovery period and the post-1993 or out-of-sample period. We find four characteristics have significant predictive power, in the cross-section, over and above that of their factors. In the out-of-sample period, five new characteristics become significant predictors. Similar results are seen for the Chen and Zhang (2010) model. Next, we find that characteristics that forecast stock returns before and after major economic events are very different. Finally, we find that the ability of characteristics to reflect economic uncertainty and sentiment changes in sign and magnitude over time- often vanishing altogether. Our results suggest that the search for a unique characteristic-based asset pricing model is unlikely to be fruitful given the secular variation in the relation between the sources of macroeconomic risks and firm-level characteristics.

Keywords: Characteristics, Anomalies, Factor Models, Risk Premia

JEL Classification: G1, G10, G14