Forecasting long-term US equity returns with a steady-state Bayesian VAR

Forecasting long-term US equity returns with a steady-state Bayesian VAR: Adapting a Bayesian VAR with steady-state priors first introduced by Villani (2009), and inspired by Shiller et al. (2021) and Lettau et al. (2008), I extend the Ferreira and Santa Clara (2011) framework to allow for reversion of equity valuations to a fair-value based on macroeconomic conditions. The proposed framework generates statistically and economically significant results by: reducing the out-of-sample root mean square forecast error of 10-year US equity returns by annualised 1.2% with respect to the Ferreira and Santa Clara (2011) method and by annualised 2.5% on a model based on simple historical average; and by generating certainty equivalent return gains for a long-term investor of annualised 2% with respect to the Ferreira and Santa Clara (2011) method and of annualised 2.75% on a model based on simple historical average. 

Modelling earnings growth to forecast long-term US equity returns: I extend Cilla (2023) framework to forecast earnings growth via a Bayesian VAR with steady-state priors. In particular, the newly introduced framework departs from the 20-year moving average assumption of Ferreira and Santa Clara (2011), and it allows for reversion of earnings growth to an equilibrium based on economic conditions. The proposed framework generates statistically and economically significant results by: reducing the out-of-sample root mean square forecast error of 10-year US equity returns by annualised 1.4% with respect to the Ferreira and Santa Clara (2011) method and by annualised 2.7% on a model based on simple historical average; and by generating certainty equivalent return gains for a long-term investor of annualised 2% with respect to the Ferreira and Santa Clara (2011) method and of annualised 2.6% on a model based on simple historical average.