On predicting ocean freight rates: a novel hybrid model of combined error evaluation and reinforcement learning

The prediction of shipping freight rates is crucial for shipping companies and related professionals, to navigate market changes, refine business strategy, and improve risk management. To improve the precision of predictions and capture the diverse underlying information in freight rates, we introduce a hybrid forecasting strategy that incorporates combined error evaluation and reinforcement learning. The model utilizes a pool comprising time series, traditional nonlinear, and deep learning models as its foundation. To ensure model efficiency and validity, we develop a combined evaluation method to assess its components. The predictions and weights of different models are then integrated to generate the hybrid prediction output. To optimize weight values, we employ reinforcement learning to dynamically update them, based on the prior prediction performance of each model, thereby further improving prediction accuracy. Empirical results based on freight rate data from three shipping sectors demonstrate that, in terms of prediction accuracy, our model outperforms earlier efforts.