The dynamics of expected returns: evidence from multi-scale time series modelling

Conventional wisdom posits that all the relevant investors' information lies at the highest possible frequency of observation, so that long-run expected returns can be mechanically inferred by a forward aggregation of short-run estimates. We reverse such logic and propose a novel framework to model and extract the dynamics of latent short-term expected returns by coherently combining the lower-frequency information embedded in multiple predictors. We show that the information cascade from low- to high-frequency levels allows to identify long-lasting effects on expected returns that cannot be captured by standard persistent ARMA processes. The empirical analysis demonstrates that the ability of the model to capture simultaneously medium- to long-term fluctuations in the dynamics of expected returns, has first order implications for forecasting and investment decisions.