Comparison of Econometric Models and Artificial Neural Networks Algorithms for the Prediction of Baltic Dry Index

• Published in: IEEE access Vol. 7; pp. 1647 - 1657
• Main Authors: Zhang, Xin, Xue, Tianyuan, Eugene Stanley, H.
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; ARIMA; ARIMA,GARCH; Artificial intelligence; Artificial neural networks; artificial neural networks(ANN); Autoregressive models; Back propagation networks; Baltic dry index prediction; Biological system modeling; BP neural network; Econometrics; ELM; Forecasting; GARCH; Global economy; Indexes; International trade; Model accuracy; Neural networks; Predictive models; RBFNN; Shipping; Time series analysis; Volatility
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2018.2884877

The shipping market, a major component of the global economy, is characterized by high risk and volatility. The Baltic dry index is an influential indicator in the world shipping market and international trade. Several studies have used a variety of techniques to generate Baltic dry index predictions. The most prominent techniques utilize either econometric or artificial intelligence computing. We compare the forecasting accuracy of two typical univariant econometric models and three artificial neural networks (ANNs)-based algorithms. We find that when using daily data, econometric forecasting models produce better one-step-ahead predictions than ANN-based algorithms. When forecasting weekly and monthly data, ANN-based algorithms produce fewer errors and a higher direction matching rate than econometric models. We also compare the predictive power of a number of different models when applied to the 2008 financial crisis and find that the generalized autoregressive conditional heteroskedasticity model and the back propagation neural network algorithm produce the best one-step-ahead and seven-steps ahead predictions, respectively.