Forecasting house price growth rates with factor models and spatio-temporal clustering
提出结合全局和聚类特定因子的模型，自动估计聚类结构，用于预测美国房价增长率，发现四个主要聚类，且该模型比仅用全局因子或无因子模型预测更准。
This paper proposes to use factor models with cluster structure to forecast growth rates of house prices in the US. We assume the presence of global and cluster-specific factors and that the clustering structure is unknown. We adopt a computational procedure that automatically estimates the number of global factors, the clustering structure and the number of clustered factors. The procedure enhances spatial clustering so that the nature of clustered factors reflects the similarity of the time series in the time domain and their spatial proximity. Considering house prices in 1975–2023, we highlight the existence of four main clusters in the US. Moreover, we show that forecasting approaches incorporating global and cluster-specific factors provide more accurate forecasts than models using only global factors and models without factors.