Dynamic Panel Data Models with Unobserved Groupings and Factor Structure

In this paper we extend the panel data model proposed by Ando and Bai (2016) by including an autoregressive component. In order to estimate the model parameters, the original estimation method was adapted in two different ways, a direct approach and an iterative approach. Of these, the iterative approach resulted in consistent estimation of the model parameters, while the direct approach had problems with convergence. By means of Monte Carlo simulation, we empirically show that independence over time of the error terms is necessary for consistent estimation of the model parameters. However, dependence over the different dependent variables of the error terms is allowed. By a similar method, it is shown that the model is robust against overspecification of the true number of lags.