A1562
Title: Detection of multiple structural breaks with latent homogeneity for large panel data
Authors: Jiatong Li - Huazhong University of Science and Technology (China) [presenting]
Degui Li - University of Macau (China)
Hongqiang Yan - Arizona State University (United States)
Abstract: This paper studies the estimation of multiple structural breaks in large panel data, allowing for heterogeneous break structures across units. A latent homogeneity framework is introduced in which cross-sectional units are partitioned into unobserved groups sharing common break locations and numbers, while heterogeneity is preserved across groups. A three-step procedure is proposed: unit-level break estimation, clustering based on Hausdorff distances between estimated break sets, and a post-clustering refinement that pools information within groups via a localized CUSUM criterion. The proposed estimator achieves exact consistency for break locations and consistently recovers the latent group structure under general conditions with both dimensions diverging. Simulation results demonstrate that the proposed method consistently recovers break locations and latent group structures when group structure is present.
