The Mechanism of Google Trends Affecting Crude Oil Price Forecasting

Searching big data combined with machine learning algorithms to improve the accuracy of financial time series forecasts has been investigated actively. This research sought to explain the internal mechanisms of predictability and expose the impact of Google Trends (GT) on the forecasting of West Texas Intermediate crude oil futures price (WTI) time series. Based on the complexity theory, the chaotic characteristics, the causal relationship of GT and WTI time series are analyzed and discussed for the first time. First, the LSTM neural network combined with the GT analysis is introduced. The prediction results reveal that the selection of GT keywords has a continuous and complicated impact. Then, the hurst exponent, fractal dimension, and Lyapunov exponent are applied to investigate the self-similarity and complexity. It indicates that the WTI and GT time series data at different time scales obey the fractal Brownian motion, and the time series have the characteristics of nonlinearity, chaotic behavior, and long-term memory. Further use of transfer entropy can effectively quantify the directionality and causality of information transfer between GT and WTI. The results show that the information flow between GT and WTI is bidirectional, asymmetric, and time-varying. This explains the weaker ability and dynamic impact of GT in improving the long-term prediction of WTI. © 2022, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.