Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)



First Advisor

R. Carter Hill


This dissertation is primary concerned on the study of semiparametric estimation approaches. In the respect of the usage of econometric analysis that is evaluating theoretical relationship, the semiparametric analysis is useful to get the flexibility of functional form. The kernel-type nonparametric methods are used for semiparametric approaches in this dissertation. The first essay focuses upon performance of various bandwidth selectors in the local linear regression method. The results indicate that the variable bandwidth selector is superior to constant bandwidth selector in the more skewed data set or complicated functional form. LSCV bandwidth selector fit well in the simple functional form. This essay also indicates that the variable bandwidth selector performs well in almost everywhere in general. The second essay is the application of local linear regression method with variable bandwidth selector to the wage equation. The challenge for the quadratic or quartic relationship between log wage and experience recently make possible to apply the semiparametric estimation method to the wage equation. The comparison of semiparametric and parametric specifications indicates that semiparametric estimation methods capture nonlinearities in the earnings profiles. Also, the analysis of wage profile using semiparametric method confirms the stylized facts of U.S. earning profiles during the 1990's. The semiparametric estimation method is applied to the qualitative response model in the third essay. The simultaneous two-stage probit model is studied using semiparametric method in the two-stage. Klein and Spady's semiparametric MLE is applied in this essay. The Monte Carlo simulation results indicate that semiparametric method performs well in the both homoscedasticity and heteroscedasticity error terms. MSE of semiparametric estimation is smaller and steadier than two-stage probit estimation.