Statistics and Its Interface

Volume 10 (2017)

Number 1

A hybrid transfer learning model for crude oil price forecasting

Pages: 119 – 130

DOI: https://dx.doi.org/10.4310/SII.2017.v10.n1.a11

Authors

Jin Xiao (Business School, Sichuan University, Chengdu, Sichan, China; and Department of Mathematics and Computer Science, University of Muenster, Germany)

Yi Hu (School of Economics and Management, University of Chinese Academy of Sciences, Beijing, China)

Yi Xiao (School of Information Management, Central China Normal University, Wuhan, China)

Lixiang Xu (Department of Mathmatics and Physics, Hefei University, Hefei, China; and Department of Mathematics and Computer Science, University of Muenster, Germany)

Shouyang Wang (Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing, China)

Abstract

Most of the existing models for oil price forecasting only use the data in the forecasted time series. This study proposes a hybrid transfer learning model (HTLM) for crude oil price forecasting. We first selectively transfer some related time series in the source domain to assist in modeling the target time series by using a transfer learning technique, and then construct the forecasting model using the analog complexing (AC) method. Further, we introduce a genetic algorithm to find the optimal match between two important parameters in HTLM. Finally, we use two main crude oil price time series—the West Texas Intermediate (WTI) and the Brent crude oil spot prices—for empirical analysis. Our results show the effectiveness and superiority of the proposed model compared with existing models.

Keywords

hybrid transfer learning model, analog complexing, genetic algorithm, crude oil price forecasting, transfer learning technique

Published 27 September 2016