Effective crude oil price forecasting using new text-based and big-data-driven model

Abstract

This study proposes a novel data-driven crude oil price prediction methodology using Google Trends and online media text mining. Convolutional neural network (CNN) is used to automatically extract text features from online crude oil news to illustrate the explanatory power of text features for crude oil price prediction. Specifically, our findings contribute to the methodological and theoretical insights for information processing, in that variational mode decomposition is used to construct useful time series indicators based on the outputs of CNN. Experimental results imply that the proposed text-based and online-big-data-based forecasting methods outperform other techniques. A total of 4837 and 3883 news headlines are collected in two cases, respectively. The mean absolute percentage error of the proposed model is 0.0571 and 0.0459 for crude oil price forecasting of two cases, respectively. Therefore, the complementary relationship between news headlines and Google Trends is beneficial in conducting considerably accurate crude oil price forecasting.