Development of Food Commodity Price Forecasting Model as an Early Warning System with a Multivariate Time Series Clustering

Fluctuations in food commodity prices have a significant impact on a country’s food security, purchasing power, and economic growth. Therefore, good governance is needed to maintain price stability, one of which is by developing a forecasting model as an early warning system. This study aims to develop a food commodity price forecasting model using Multivariate Time Series Clustering (MTSClust) and Vector Autoregressive Imputation Method with Moving Average (VAR-IMMA) approaches for food commodities in the Indonesian region. The data used in this study consisted of daily prices of 13 commodities from 103 districts/cities in Indonesia. Data analysis was conducted in several stages, namely VAR modeling, K-means Euclidean clustering, profiling, and forecasting. The results show that 103 sample districts/cities across Indonesia can be grouped into four types of regions based on food price movement patterns. There are homogeneous islands such as Maluku where the sample district/city are in the same cluster, but there are also heterogeneous islands such as Kalimantan and Papua with their four clusters. The forecasting evaluation results show good accuracy with Root Mean Square Error (RMSE) scores below IDR 1000.00 in most cases, which is equivalent to Mean Absolute Percentage Error (MAPE) scores below 10%. However, two commodities, namely cayenne pepper and red chili, need more attention due to relatively high RMSE and MAPE scores, although not exceeding 30% MAPE in most cases. These results show that the MTSClust and VAR-IMMA approaches are accurate in forecasting food commodity prices, although further research is needed for the two chili commodities.