Equity premium prediction and structural breaks

A Bayesian autoregressive model that allows for multiple structural breaks outperforms the historical average, which has proven so successful, in a statistically and economically significant way for mean‐variance investors when forecasting the equity premium. A range of autoregressive models that do not allow for breaks or do so in an ad hoc way fail to outperform the historical average. The Bayesian model estimates three breaks that occur in 1929, 1940, and 1971 corresponding to major events that drive the shifts in the underlying distribution of the premium. Allowing for breaks over the forecast horizon further improves the forecasting power.