Panel Data Dynamics and Measurement Errors: GMM Bias, IV Validity and Model Fit – A Monte Carlo Study
Erik Biørn and Xuehui Han
An autoregressive fixed effects panel data equation in error-ridden endogenous and exogenous variables, with finite memory of disturbances, latent regressors and measurement errors is considered. Finite sample properties of GMM estimators are explored by Monte Carlo (MC) simulations. Two kinds of estimators are compared with respect to bias, instrument (IV) validity and model fit: equation in differences/IVs levels, equation in levels/IVs in differences. We discuss the impact on estimators’ bias and other properties of their distributions of changes in the signal-noise variance ratio, the length of the signal and noise memory, the strength of autocorrelation, the size of the IV set, and the panel length. Finally, some practical guidelines are provided.