Abstract
This paper investigates whether macroeconomic variables can improve the predictability of equity risk premia. Ex ante forecasts of excess returns are generated recursively from both linear regression analysis and nonlinear neural networks. Empirical results suggest that these forecasts are superior to the random walk predicator at almost all horizons in the U.S., Japanese, British and German stock markets. However, there does not appear to be a significant difference between linear and nonlinear forecasts.