A Study of Multivariate Time Series Clustering Based on Vector Autoregressive with Exogenous Models for Forecasting Cooking Oil Prices in Indonesia

Abstract

The fluctuations in cooking oil prices in 77 regencies and cities included in the Strategic Food Price Information Center data show varying levels of volatility over time. This phenomenon presents a challenge for forecasting cooking oil prices in Indonesia in the future. Consequently, a method capable of estimating these prices is required. One such approach involves Multivariate Time Series Clustering (MTSCLUST) method based on the Vector Autoregressive with Exogenous (VARX) model. The MTSClust method is employed to map spatial time series pattern interdependencies across regions, while the VARX model is utilized for its ability to capture the dynamics of intervariable relationships influenced by external factors. This study aims to examine the application of the MTSClust method based on the VARX model in forecasting cooking oil prices in Indonesia in the future. Cluster formation results with K-Means with the Chebyshev distance metric indicate that the 77 regencies and cities in Indonesia can be grouped into three clusters, achieving excellent forecasting accuracy with MAPE values below 10%. Cluster 3 became the best cluster using the VARX (24,1) model, demonstrating optimal performance with a MAPE value of 0.228%. The findings of this study confirm that the MTSClust model with VARX integration is effective for forecasting future cooking oil prices.