Forecasting crude oil prices using discrete wavelet transform with autoregressive integrated moving average and Least Square Support Vector Machine combination approach

Abstract

In this paper, a hybrid time series forecasting approach is proposed consisting of wavelet transform as the data decomposition method with Autoregressive Integrated Moving Average (ARIMA) and Least Square Support Vector Machine (LSSVM) combination as the forecasting method to enhance the accuracy in forecasting the crude oil spot prices (COSP) series. In brief, the original COSP is divided into a more stable constitutive series using discrete wavelet transform (DWT). These respective sub-series are then forecasted using ARIMA and LSSVM combination method, and lastly, all forecasted components are combined back together to acquire the original forecasted series. The datasets consist of monthly COSP series from West Texas Intermediate (WTI) and Brent North Sea (Brent). To evaluate the effectiveness of the proposed approach, several comparisons are made with the single forecasting approaches, a hybrid forecasting approach and also some existing forecasting approaches that utilize COSP series as the dataset by comparing the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) acquired. From the results, the proposed approach has managed to outperform the other approaches with smaller MAE and RMSE values which signify better forecasting accuracy. Ultimately, the study proves that the integration of data decomposition with forecasting combination method could increase the accuracy of COSP series forecasting.