Prediction of Crude Oil Prices in a Mixed-Frequency Data-Rich Condition : New Evidence from the LASSO-MIDAS Model

In a data-rich environment, few studies use many predictors at different frequencies to forecast crude oil returns. This paper proposes a new model, LASSO-MIDAS, which combines LASSO (the least absolute shrinkage and selection operator) and MIDAS (mixed data sampling model) to predict crude oil returns with many mixed-frequency predictors (e.g., monthly, weekly, and daily). The empirical results demonstrate that the forecasting performance of the LASSO-MIDAS model is more successful than that of various competing models, including the Ridge-MIDAS, PCA-MIDAS, SPCA-MIDAS and Group-LASSO-MIDAS models. Moreover, our findings are reliable for a set of robustness checks, such as different estimated windows, different regularization parameters, and alternative predictors. Furthermore, the more precise forecasting ability of the LASSO-MIDAS model can obtain a much larger economic value than other forecasting models for a mean-variance investor. Overall, our extended study provides strong evidence that mixed-frequency indicators can enhance forecasting accuracy more than single-frequency indicators in the crude oil market