Research On The Parameter Estimation Problem Of Panel Data Models

In statistics and econometrics, the term panel data refers to two dimensionaldata. Panel data contains observations on two phenomena observed over multipletime periods for each individual. For example, in Balestra and Nerlove (1966)，panel data on 36 U. S. states over a 13 year period were used in the analysis.Time series and cross-sectional data are special cases of panel data that are inone-dimension only. Panel data are sometimes treated as cross-sections over timeor pooled cross-section time-series data.Empirical research in economics has been enriched by the availability of awealth of panel data. These allow us to construct and test more realistic paneldata models. So, theoretical research on panel data models has been an importantbranch of modern econometrics, called panel data econometrics, of which researchon parameter estimation problem of dynamic panel data model and dynamic bi-nary panel data model is the hot and difficult problems in today’s internationaleconometric studies. The so-called dynamic models is that models contain laggeddependent variables, i.e. lagged dependent variables appear as explanatory vari-abler. The regressors of the dynamic panel data model and dynamic binary paneldata model are not strict exogenous variables, so the maximum likelihood esti-mator (MLE) is biased and inconsistent. This dissertation aims to propose a biascorrection method based on the maximum likelihood estimator (MLE), and thebias correction estimator is asymptotically unbiased and consistent.In this dissertation, we first study the parameter estimation problem of dy-namic panel data with fixed effects, and a由ust the maximum likelihood estima-torn to result in asymptotically unbiased and consistent estimators. Secondly,we research the hot problem of nonlinear panel data models--estimating thedynamic binary panel data models with fixed effects. A bias correction estimatorbased on MLE is proposed, which is asymptotically unbiased and consistent. Oneof the main advantages of the bias correction method over other methods for esti- mating dynamic binary panel data models is its generality. For instance, it is notrestricted to the logistic case. The bias correction estimator is generally appli-cable and it has the same asymptotic properties regardless of the distribution ofthe errors. Finally, for dynamic panel data model and dynamic binary panel datamodel, Monte Carlo simulation studies are respectively conducted to evaluate thefinite sample properties of iterative bootstrap bias correction estimator.