Forecasting with model selection or model averaging: a case study for monthly container port throughput

An accurate short-term prediction of time series data is critical to operational decision-making. While most forecasts are made based on one selected model according to certain criteria, there are developments that harness the advantages of different models by combining them together in the prediction process. Following on from existing work, this paper applies six model selection criteria and six model averaging (MA) criteria to a structural change vector Autoregressive model, and compares them in terms of both the theoretical background and empirical results. A case study of the monthly container port throughput forecasting for two competing ports shows that, in general, the model averaging methods perform better than the model selection methods. In particular, the leave-subject-out cross-validation MA method is the best in the sense of achieving the lowest average of mean-squared forecast errors.