A2004
Title: Bayesian SARIMA modelling with a spatial error structure for electricity demand in Japan
Authors: Hanzhi Yu - Aoyama Gakuin University (Japan) [presenting]
Haruhisa Nishino - Aoyama Gakuin University (Japan)
Abstract: Regional electricity demand in Japan exhibits strong seasonality and cross-regional dependence, requiring forecasting models that jointly account for temporal and spatial structures. A Bayesian seasonal autoregressive integrated moving average (SARIMA) framework is developed for regional electricity demand, and alternative spatial specifications are compared using quarterly and monthly regional data. Predictive performance is evaluated using WAIC and Pareto-smoothed importance sampling leave-one-out cross-validation. The findings indicate that combining seasonal dynamics with a spatial error structure improves predictive performance and provides a computationally efficient way to capture spatial dependence. In the quarterly data model, an electricity price proxy, defined as sales revenue divided by electricity sales volume, is also incorporated into the model and analysed. The Great East Japan Earthquake appears to have a largely temporary effect on electricity demand, whereas electricity market liberalization is interpreted as a regime shift. Overall, jointly modelling seasonal patterns and spatial error dependence can be useful for regional electricity demand forecasting in Japan.
