Bicluster Analysis of Heterogeneous Panel Data Via M-Estimation

Weijie Cui,Yong Li
DOI: https://doi.org/10.3390/math11102333
IF: 2.4
2023-01-01
Mathematics
Abstract:This paper investigates the latent block structure in the heterogeneous panel data model. It is assumed that the regression coefficients have group structures across individuals and structural breaks over time, where change points can cause changes to the group structures and structural breaks can vary between subgroups. To recover the latent block structure, we propose a robust biclustering approach that utilizes M-estimation and concave fused penalties. An algorithm based on local quadratic approximation is developed to optimize the objective function, which is more compact and efficient than the ADMM algorithm. Moreover, we establish the oracle property of the penalized M-estimators and prove that the proposed estimator recovers the latent block structure with a probability approaching one. Finally, simulation studies on multiple datasets demonstrate the good finite sample performance of the proposed estimators.
What problem does this paper attempt to address?