Authors: Theodore Syriopoulos, Michael Tsatsaronis, Ioannis Karamanos, Mikaela Oikonomopoulou

Title: Support Vector Machine (SVM) Assessment in Shipping Freights Forecasting

Abstract

Robust statistical predictions of critical time-series variables are a key concern in finance and economics. In shipping finance, forecasting shipping freights and modeling the dynamics of shipping markets is of vital importance. This study investigates an innovative approach to forecasting freights n different routes and subsectors, incorporating and testing the feasibility and predictive power of a competitive learning algorithm, the support vector machine (SVM) model. For this purpose, a diverge set of econometric models from the ARIMA family are also considered. The performance of each model is evaluated for both training and forecasting data, according to the Root–Mean–Square Error (RMSE), Mean Absolute Error (MAE) and correlation coefficient (r) that are widely used to evaluating time-series forecasting results. The experimental results obtained indicate that the SVM model outperforms the standard econometric models, implying that SVM is a promising approach to shipping freights forecasting. The application of SVM in time-series forecasting is relatively new. However, to the authors’ best knowledge, the incorporation of SVM in shipping markets price forecasting is undertaken for the first time; hence, it remains an innovative and challenging empirical exercise. The empirical results demonstrate that the SVM model can be a promising alternative approach to time-series predictions; furthermore, it can be implemented in shipping risk management and financing decisions.