Asymptotics and Bootstrap for Transformed Panel Data Regressions1

Liangjun Su,Zhenlin Yang

2008-01-01

Abstract:This paper investigates the asymptotic properties of quasi-maximum likelihood es- timators for transformed random effects models where both the response and (some of) the covariates are subject to transformations for inducing normality, flexible functional form, homoscedasticity, and simple model structure. We develop a quasi maximum likelihood-type procedure for model estimation and inference. We prove the consistency and asymptotic normality of the parameter estimates, and propose a simple bootstrap procedure that leads to a robust estimate of the variance-covariance matrix. Monte Carlo results reveal that these estimates perform well in finite samples, and that the gains by using bootstrap procedure for inference can be enormous.