Homogeneity Pursuit of Panel Data with Sparsity
In panel data analysis, studying individual attributes is important for both methodological development and practical applications. Homogeneity in individual attributes and sparsity in coefficients are therefore important topics, as they help uncover relationships among individuals and identify the truly relevant covariates. However, many existing approaches do not explicitly exploit the potential grouping information contained in separately fitted individual-level penalized estimators to identify latent groups. To address this gap, we propose a hierarchical group recovery approach based on sparse initial estimation to jointly identify latent groups and sparse parameter structures in panel data. We first introduce a basic version that can efficiently identify homogeneity and sparsity. Based on this basic version, we further introduce a local search step to improve the performance of our method. Compared with existing methods, our approach shows the best overall performance when jointly considering estimation accuracy and computational efficiency. The simulation results demonstrate the effectiveness and advantages of our approach over existing approaches. We further apply the proposed approach to a real dataset to show its practical applicability.