Monthly ship price forecasting based on multivariate variational mode decomposition

Abstract

Accurate and reliable ship price forecasts can assist shipping firms, investors, and other participants to withstand risks and make profits in a highly volatile shipping market. Considering the nonlinear behavior and sophisticated interrelationship of the shipping market, in this paper, a novel multiscale and multivariable methodology based on multivariate variational mode decomposition (MVMD) and machine learning (ML) algorithms is proposed for forecasting monthly newbuilding ship price (NSP), secondhand ship price (SSP), and ship scrap value (SSV). The proposed methodology involves three main modules: (1) data decomposition by MVMD; (2) mode forecasting using ML algorithms, including multi-layer perceptron, support vector regression, long short-term memory, and gated recurrent unit; and (3) ensemble forecasts via simple addition. The novelty of this paper lies in the employment of MVMD, which is capable of capturing complicated multiscale relationships among the NSP, SSP, and SSV data by extracting multiple frequency-aligned oscillatory modes. This will largely help in the multivariable forecasting of ship prices at each timescale. With Capesize bulker and VLCC tanker as study samples, the empirical results show that the novel methodology outperforms other considered benchmark models in terms of both level and directional forecasting accuracy. This suggests that the developed multiscale and multivariable methodology is a promising alternative for shipping market analysis and forecasting.