Multifactors risk loadings and abnormal returns under uncertainty and learning

This paper considers an environment where investors have limited knowledge of true systematic risks and therefore continuously re-estimate the forecasting model that they use to form expectations. Based on a parsimonious specification with learning and no conditioning information, I extract time-varying factor loadings, pricing errors, and a measure of pricing uncertainty for the Fama-French three-factor model. Estimated parameters display significant fluctuations over time, both short-run and long-term, along patterns that vary across industry portfolios. Besides being markedly variable across portfolios and over time, abnormal returns and risk loadings also display strong systematic correlations with market conditions and business-cycle developments. Overall, the estimates convey the idea that over the past two decades stocks have experienced a pervasive increase in the variability of their exposure to fundamental risks.