To Pool or Not to Pool: A Partially Heterogeneous Framework

This paper proposes a partially heterogeneous framework for the analysis of panel data with fixed T , based on the concept of "partitional clustering". In particular, the population of cross-sectional units is grouped into clusters, such that parameter homogeneity is maintained only within clusters. To de- termine the (unknown) number of clusters we propose an information-based criterion, which, as we show, is strongly consistent - i.e. it selects the true number of clusters with probability one as N approaches infinity. Simulation experiments show that the proposed criterion performs well even with moderate N and the resulting parameter estimates are close to the true values. We apply the method in a panel data set of commercial banks in the US and we find four clusters, with significant differences in the slope parameters across clusters.