Time Series Clustering in High Dimensional Cointegration Analysis: The Case of African Swine Fever in China

Time series data have been extensively utilized in agricultural price analysis, with the Vector Auto-Regressive (VAR) and Vector Error Correction Model (VECM) being foundational tools. Over the past three decades, the availability of disaggregated agricultural commodity price data has increased, resulting in high-dimensional datasets. The efficacy of VECM and Johansen’s maximum likelihood test diminishes with increased dimensionality due to exponential growth in the required time series length, implying difficulty in extracting cointegrating relationships in high-dimensional data. This article addresses this challenge by employing time series clustering to reduce data dimensionality. Clusters are formed based on price similarity, dynamically adjusted for specified time period using hierarchical clustering with dynamic time warping. With clustered time series, we extract the mean price of each cluster and apply Johansen’s framework to estimate cointegration relationships. Applied to the Chinese hog market before and after the 2018 African Swine Fever outbreak, we show that the cointegrating relationship has changed suggesting less inter-provincial trade. The study identifies clusters based on price similarity and shows the advantages of this method compared to traditional geographical clustering.