Forecasting returns in the crude oil and heating oil markets via regularized VAR models

This study pioneers the application of regularized vector autoregressive (VAR) models in return forecasting research and explores their effectiveness in predicting the spot and futures returns of crude oil and heating oil. Our findings demonstrate the efficacy of regularized VAR models, especially the superior predictive performance of the VAR-elastic model with a small lag order and the VARX-elastic model (the model that incorporates predictors as exogenous variables into the VAR-elastic framework) with a large lag order, in predicting the futures and spot returns of crude oil and heating oil. Furthermore, the predictive power of the regularized VAR models for crude oil and heating oil returns is primarily concentrated during periods of economic recessions. Finally, mean-variance investors operating within crude oil and heating oil markets can achieve considerable utility gains by employing regularized VAR models, particularly the VAR-elastic model with a small lag order and the VARX-elastic model with a large lag order.