Predicting equity premium by conditioning on macroeconomic variables: A prediction selection strategy using the price of crude oil

A growing number of studies use crude oil price measures possibly in combination with other financial and macroeconomic variables, such as the dividend yield and earnings-to-price ratio as predictors in econometric models to evaluate whether they improve the accuracy of equity premium point predictions out-of-sample. In this study, we suggest relying on widely used crude oil price measures not as regressors in our predictive model but rather using them to devise a prediction selection strategy. Particularly, we use the crude oil price measure of interest to select between predictions produced under the benchmark and the predictive model augmented with a financial or macroeconomic variable. Using this prediction selection strategy, we succeed in obtaining point prediction accuracy improvements close to 10% relative to the benchmark and commonly used alternatives.