Monte Carlo Evaluations of Common State Dependence Estimators

This thesis represents an attempt to provide a deeper knowledge of the finite sample properties of some econometric methods used to estimate the magnitude of state dependence in binary choice dynamic panel models. These models are often applied in labor economics. The models I evaluate are the Heckman method, Wooldridge method and the linear probability model using Arellano-Bond instruments. By carefully designing appropriate Monte Carlo experiments I test the models' performance under different assumptions and different distributions of the error term, individual-specific fixed effects and explanatory variables. The results indicate that the Heckman method is the most precise estimator in most cases, followed by the linear probability model. The Wooldridge method, while seldom the most accurate, is shown to be robust to violated assumptions. The linear probability model breaks down when the process includes an age-trended variable and the Heckman method breaks down when the explanatory variable is correlated with the individual-specific fixed effects. In most cases the three estimation methods display satisfactory performance. There are only modest performance gains from increasing the number of observed time periods