​

Abstract: The Chinese commodity futures market has become an essential component of the global maritime transport system as international trade continues to expand. The China Containerised Freight Index (CCFI) serves as a valuable indicator of the maritime industry's health and is highly sensitive to fluctuations in the Chinese commodity futures market. However, there is a lack of research utilizing Chinese commodity futures prices as predictors for the CCFI. This study analyzes a dataset comprising 29,308 observations collected daily from March 24, 2017, to May 27, 2022. We conduct a comparative analysis of CCFI prediction using Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and a hybrid CNN-LSTM model. The CNN-LSTM model effectively identifies nonlinear features within CCFI data and captures the long term dependencies of the index over time, as evidenced by our results. This model outperforms the individual CNN and LSTM approaches and demonstrates high adaptability to fluctuations arising from random sample selection, data frequency, and structural discontinuities within the sample population. This study highlights the potential of machine learning methods for forecasting shipping indices, thereby enhancing understanding of the relationship between the shipping industry and financial markets. The findings provide logistics companies, shipping organizations, and governments with robust risk management and decision-support tools.

Keywords: CCFI forecast, futures market, machine learning, convolution neural network, long and short-term memory