A hybrid combination approach to forecast freight rates volatility

The aim of this paper is to investigate the performance of machine learning algorithms along with traditional GARCH and GARCH-MIDAS models in forecasting volatility of dry bulk shipping freight rates, known as one of the most volatile asset classes. In doing so, we introduce a new market tightness index, capturing physical constraints in shipping markets as an explanatory variable. The results suggest that significant incremental information can be extracted by Machine Learning algorithms from additional volatility predictors with minimal noise fitting, if regularization is applied. However, traditional GARCH models perform better in capturing the long-term persistence of the volatility. Therefore, a novel hybrid ensemble stacking algorithm that combines GARCH models and tree-based algorithms is proposed. This hybrid model, which utilizes exogenous predictors and the GARCH-MIDAS specification with the marked tightness index, produces accurate and robust out-of-sample volatility forecasts over a range of time horizons, from one day to one month.