Macroeconomic Extrapolation, Machine Learning, and Equity Risk Premium Forecast

We present a macroeconomic trend extrapolation approach that utilizes all economic fundamentals of different time periods simultaneously in the aggregate market. We demonstrate that the trend-pooling method statistically and economically outperforms the historical average that assumes a constant equity risk premium, as well as Rapach, Strauss, and Zhou’s (2010) simple pooling method that ignores the historical information in macroeconomics. We further find that extrapolating via machine learning techniques generates monthly out-of-sample R2 statistics as high as 4%. Extrapolating economic fundamentals with a grid of moving averages more closely tracks important macroeconomic fluctuations and more effectively regulates the forecast variability, thereby generating superior and robust forecasting gains consistently over time