Predicting crude oil prices using UAR, LSTM and hybrid models

Crude oil price time series data exhibit high complexity, strong correlation and long-term dependence. The accurate prediction of crude oil price trend has important significance for formulating macroeconomic policy, national energy policy and sustainable development policy. In this study, 228 monthly WTI crude oil price datasets from January 2001 to December 2019 were used. To estimate crude oil price time series, both the UAR (uncertain autoregressive) model and LSTM (long short-term memory) model are employed. As to get a better estimation effect, a hybrid model based on UAR and LSTM models is used for prediction. By comparing the NMSE (normalized mean squared error), NRMSE (normalized square root error), and NMAE (normalized mean absolute error) results of the three methods, the hybrid model has been proven to have the best prediction performance and the lowest prediction error. Specifically, the NMSE values of the UAR, LSTM, and UAR-LSTM models are 3.0419, 9.4696, and 1.1351, respectively, which further confirms the significant advantage of the hybrid model in terms of accuracy. The results of all applied models are consistent, which indicates the success of crude oil price prediction.