Machine learning and oil price point and density forecasting

The purpose of this paper is to explore machine learning techniques forecast the oil price. In era big data, we investigate whether new automated tools can improve over traditional approaches in terms accuracy. Oil price point and density forecasts are built from 23 methods, including regression trees (random forest, quantile xgboost), regularization procedures (elastic net, lasso, ridge), standard econometric models combinations, besides structural factor model Schwartz Smith (2000). database contains 315 macroeconomic financial variables, used build high-dimensional models. To evaluate predictive power each method, an extensive pseudo out-of-sample forecasting exercise built, monthly quarterly frequencies, with horizons one month up five years. Overall, results indicate a good performance methods short-run. Up six months, lasso-based models, future prices, VECM Schwartz–Smith provide best forecasts. At longer horizons, combinations also become relevant. several cases, accuracy gains respect random walk statistically significant reach two-digit figures, percentage terms, using R2 statistic; expressive achievement compared previous literature.