Forecasting Using Heterogeneous Panels with Cross-Sectional Dependence

In this paper, we focus on forecasting heterogeneous panels in presence of cross-sectional dependence in terms of both spatial error dependence and common factors. We propose two main approaches to estimate the factor structure, one using the residuals (Residuals Based Approach", RBA) while the second using a panel of some variables (Auxiliary Variables Approach", AVA) to extract the factors. Small sample properties of the methods proposed is investigated through Monte Carlo simulation exercises and used in an application to predict house price inflation in OECD countries.