Regression clustering for panel-data models with fixed effects

In this article, we describe the xtregcluster command, which implements the panel regression clustering approach developed by Sarafidis and Weber (2015, Oxford Bulletin of Economics and Statistics 77: 274–296). The method classifies individuals into clusters, so that within each cluster, the slope parameters are homogeneous and all intracluster heterogeneity is due to the standard two-way error-components structure. Because the clusters are heterogeneous, they do not share common parameters. The number of clusters and the optimal partition are determined by the clustering solution, which minimizes the total residual sum of squares of the model subject to a penalty function that strictly increases in the number of clusters. The method is available for linear short panel-data models and useful for exploring heterogeneity in the slope parameters when there is no a priori knowledge about parameter structures. It is also useful for empirically evaluating whether any normative classifications are justifiable from a statistical point of view.