A secondary decomposition-ensemble methodology for forecasting natural gas prices using multisource data

In order to improve the accuracy of natural gas price forecasting, this study proposes a secondary decomposition-ensemble methodology using multiple helpful predictors, namely energy data and search engine data (SED). Four main steps are involved: primary decomposition, secondary decomposition of the component group with the highest complexity, component prediction and ensemble. The novelties include the use of a new single model, multisource data and multivariate secondary decomposition. Using data of two main natural gas prices, an empirical study is conducted to illustrate and verify the proposed methodology. The results suggest that models with multisource data (both energy data and SED) can obtain better forecasting performance than those with single source data (energy data or SED). In particular, SED can be more helpful predictors of natural gas prices than energy data. Compared with the primary decomposition-ensemble model, the corresponding model with secondary decomposition can achieve higher predictive accuracy and greater robustness.