Network-Based Clustering for Varying Coefficient Panel Data Models
提出一种变系数面板数据模型，通过三重局部化估计个体系数函数，再用社区检测识别未知的组结构，并给出两阶段估计方法达到最优收敛速度，适用于经济学等领域的面板数据分析。
In this article, we introduce a novel varying-coefficient panel-data model with locally stationary regressors and unknown group structure, in which the number of groups and the group membership are left unspecified. We develop a triple-localization approach to estimate the unknown subject-specific coefficient functions and then identify the latent group structure via community detection. To improve the efficiency of the first-stage estimator, we further propose a two-stage estimation method that enables the estimator to achieve optimal rates of convergence. In the theoretical part of the article, we derive the asymptotic theory of the resultant estimators. In the empirical part, we present several simulated examples together with an analysis of real data to illustrate the finite-sample performance of the proposed method.