Latent group structure in linear panel data models with endogenous regressors

We show that direct appli-cations of the Kmeans algorithm to the GMM objective functions do not yield unique estimates. Nonetheless, various alternative approaches are available, and they yield consistent estimators.

We examine the difference in the conditions required for each of the estimation methods. Monte Carlo simulations are conducted to compare these approaches in terms of their classification accuracy.

As an empirical application, we apply our approaches to revisit the relationship between democracy and income.