Assessing ARIMA Model Performance in Hierarchical Time Series Forecasting of Tourist Arrivals: The Role of Data Normalization and Bottom-Up Strategy

Abstract

This research evaluates the performance of the ARIMA method in forecasting hierarchical time series data of tourist arrivals in Australia from 1998 to 2016 using a bottom-up strategy. A comparative analysis is conducted between the predicted results and the actual data for both short-term and long-term periods, as well as with various normalization methods for each hierarchical level. The study concludes that ARIMA generally performs better in short-term forecasting at a hierarchical level. However, the evaluation results indicate that SMAPE (Symmetric Mean Absolute Percentage Error) values fluctuate across different forecasting periods, influenced by prediction data generated from various ARIMA models. This study does not determine whether one normalization method is superior to another, as the evaluation results show no significant differences. Nevertheless, this research provides insights into the effectiveness of hierarchical time series forecasting using the ARIMA method and a bottom-up strategy at each hierarchical level for both short-term and long-term periods. It also assesses the performance of various normalization methods used.