Maximum Likelihood Estimation of Fixed Effects Dynamic Panel Data Models Covering Short Time Periods

A transformed likelihood approach is suggested to estimate fixed effects dynamic panel data models. Conditions on the data generating process of the exogenous variables are given to get around the issue of ?incidental parameters?. The maximum likelihood (MLE) and minimum distance estimator (MDE) are suggested. Both estimators are shown to be consistent and asymptotically more efficient than the instrumental variable (IV) or generalised method of moment (GMM) estimators. A Hausman-type specification test is suggested to test the fixed versus random effects specification or conditions on the data-generating process of the exogenous variables. Monte Carlo studies are conducted to evaluate the finite sample properties of the MLE, MDE, IV and GMM. It is shown that the likelihood approach appears to dominate the GMM approach both in terms of the bias or root mean squares error of the estimators and the size and power of the test statistics.