Provincial Clustering using GARCH-based Chili Price Volatility Features

Abstract

Bird's eye chili is a strategic food commodity in Indonesia whose prices are highly susceptible to interregional fluctuations due to differences in distribution systems and supply chain conditions. These fluctuations often occur over short time horizons, making analyses based on monthly or annual data less capable of capturing short-term price spikes that have the greatest impact on consumers' purchasing power and price stabilization policies. Previous studies have generally clustered regions based on nominal or average prices, which do not adequately represent the dynamics of daily price movements. This study aims to cluster Indonesian provinces according to the volatility characteristics of bird's eye chili prices by proposing a clustering approach that utilizes GARCH(1,1)-based conditional volatility features to represent daily price dynamics, combined with the K-Means algorithm for cluster formation. Daily bird's eye chili price data from 34 provinces covering the period from April 2024 to April 2026 were obtained from the National Strategic Food Price Information Center (PIHPS). The price data were transformed into daily returns and modeled using GARCH(1,1) to estimate the average conditional volatility of each province, which was subsequently used as the clustering feature. The optimal number of clusters was determined using the Elbow Method and the Silhouette Score. The evaluation results identified five as the optimal number of clusters, achieving a silhouette score of 0.65 and classifying the 34 provinces into five volatility categories: very high, high, moderate, low, and very low. The findings reveal that provinces with higher price levels do not necessarily belong to the highest volatility cluster, indicating that a volatility-based approach provides additional insights into price dynamics beyond those captured by nominal price-based clustering. These results can support regional food price volatility monitoring and serve as a reference for developing data-driven decision support systems for food price management.