Persistence, Signal-Noise Pattern and Heterogeneity in Panel Data: With an Application to the Impact of Foreign Direct Investment on GDP
Erik Biørn and Xuehui Han
GMM estimation of autoregressive panel data equations in error-ridden variables when the noise has memory, is considered. The impact of variation in the memory length in signal and noise spread and in the degree of individual heterogeneity are discussed with respect to ﬁnite sample bias, using Monte Carlo simulations. Also explored are also the impact of the strength of autocorrelation and the size of the IV set. GMM procedures using IVs in diﬀerences on equations in levels, in general perform better in small samples than procedures using IVs in levels on equations in diﬀerences. A case-study of the impact of Foreign Direct Investment (FDI) on GDP, inter alia, contrasting the manufacturing and the service sector, based on country panel data supplements the simulation results.