A new crude oil futures forecasting method based on fusing quadratic forecasting with residual forecasting

As an essential energy commodity, crude oil plays a vital role in the global economy. Accurate forecasting of crude oil is a critical guide in determining economic policies. This study proposes a hybrid forecasting model SEL-FR-XGBoost-GNB based on fusing quadratic forecasting with residual forecasting to achieve high accuracy in forecasting crude oil futures. The model-building process includes three stages. In stage I, the crude oil futures series are predicted using SVM, ELM, and LSTM models, respectively. In stage II, the prediction results of the above three single models are first reconstructed using FR. And then, the XGBoost method is used to make a secondary prediction of the crude oil futures series. In stage III, the residual sequences of the second prediction results are trained and predicted using the GNB method. The residual prediction result and the second prediction result are added as the final prediction result. Through the forecasting study of OPEC's historical crude oil futures series, the following conclusions can be drawn: (a) the proposed FR-XGBoost-based quadratic forecasting method can make the single model SVM, ELM, and LSTM form a complementary advantage and effectively improve crude oil futures forecasting accuracy; (b) extracting the positive and negative attributes of the residual sequence and transforming the regression prediction problem into a classification prediction problem can significantly improve the predictability of the residual sequence; (c) the proposed GNB residual sequence prediction method helps to improve the performance of the hybrid model; and (d) the proposed hybrid prediction model SEL-FR-XGBoost-GNB has the best performance among the 16 general models and 4 recent existing models. (C) 2022 Elsevier Inc. All rights reserved.