Asset pricing model uncertainty

This paper provides a unified calendar-time portfolio methodology for assessing whether returns following an event are abnormal which efficiently handles asset pricing model uncertainty and allows for time-varying alpha and factor exposures. The approach disciplines researchers' use of asset pricing factors and assigns a probability measure to the appropriateness of (dynamically) selecting a single model that best approximates the true factor structure or whether model averaging across an asset pricing universe is desired. It is applied to the long-horizon effect of dividend initiations and resumptions in the 1980 to 2015 period. Resulting post-announcement conditional abnormal returns are generally significant, statistically and economically, which contrasts recent evidence, and exhibits a break in mean from positive until the mid-1990s and negative onwards. We document substantial time-variation in the dimensionality and composition of the factor structure in expected returns, which goes beyond what captured by conditional versions of the CAPM and Fama-French specifications. This also generalizes to a large panel of 202 characteristics-sorted portfolios.