Forecasting Container Throughput: A Comparison of Time-Series Methods

Abstract

International maritime transport has been the backbone of globalized trade, the manufacturing supply chain, and the global economy for a long time. A vast literature on transportation economics has argued that containerization was the major change in 20th-century transportation technology responsible for the acceleration of the globalization of the world economy since the 1960s. Ports are a very important component of global maritime trade and provide a crucial interface between land and sea. Port throughput is one of the main management objectives of port enterprises as well as one of the comprehensive indicators of port performance, economic development, and trend of port areas. If port infrastructures are under capacity the port might lose its competitive positions and if it exceeds the market demand the equipment will not be fully utilized resulting in loss of productivity. Thus, throughput forecast can be instrumental for adequate port development and efficient port operation.

Different available forecasting techniques can be broadly classified into quantitative and qualitative techniques. Qualitative methods of forecasting are based on judgments, opinions, intuitions, and experiences thus are subjective in nature and are prone to different cognitive biases. While quantitative methods of forecasting are based on mathematical models and are objectives on nature. This thesis aims to compare the performance of different quantitative methods namely ARIMA, SARIMA, HWES, Prophet, and Combined methods. Monthly container throughput data from January 2010 to December 2019 of three major ports, Shanghai, Busan, and Nagoya are used for analysis and comparison of different models. For comparing the performances of the models used, four error metrics are used namely, MAE, MSE, RMSE, and MAPE. Experimental results are then discussed and analyzed to find the best performing model. From the experiment combined method using the weighted average seems to outperform all the other models. The best performing model is then used to forecast container throughput for the next two years for all three ports. The forecast presented can be used as a guideline by the port authorities for planning and managing future demand.

International maritime transport has been the backbone of globalized trade, the manufacturing supply chain, and the global economy for a long time. A vast literature on transportation economics has argued that containerization was the major change in 20th-century transportation technology responsible for the acceleration of the globalization of the world economy since the 1960s. Ports are a very important component of global maritime trade and provide a crucial interface between land and sea. Port throughput is one of the main management objectives of port enterprises as well as one of the comprehensive indicators of port performance, economic development, and trend of port areas. If port infrastructures are under capacity the port might lose its competitive positions and if it exceeds the market demand the equipment will not be fully utilized resulting in loss of productivity. Thus, throughput forecast can be instrumental for adequate port development and efficient port operation.

Different available forecasting techniques can be broadly classified into quantitative and qualitative techniques. Qualitative methods of forecasting are based on judgments, opinions, intuitions, and experiences thus are subjective in nature and are prone to different cognitive biases. While quantitative methods of forecasting are based on mathematical models and are objectives on nature. This thesis aims to compare the performance of different quantitative methods namely ARIMA, SARIMA, HWES, Prophet, and Combined methods. Monthly container throughput data from January 2010 to December 2019 of three major ports, Shanghai, Busan, and Nagoya are used for analysis and comparison of different models. For comparing the performances of the models used, four error metrics are used namely, MAE, MSE, RMSE, and MAPE. Experimental results are then discussed and analyzed to find the best performing model. From the experiment combined method using the weighted average seems to outperform all the other models. The best performing model is then used to forecast container throughput for the next two years for all three ports. The forecast presented can be used as a guideline by the port authorities for planning and managing future demand.