Parametric and Semiparametric Binary Choice Estimators - Evidence from Monte Carlo Studies

The following thesis compares the performance of several parametric and semiparametric estimators in binary choice models using the method of Monte Carlo studies. Particularly, the thesis compares estimators of the parametric linear probability-, logit- and probit model, a model derived from Cauchy distributed errors (cauchit model) as well as the estimator proposed by Klein and Spady (KS) and the local likelihood logit estimator by Frölich (LLL), which are of the semiparametric class. Furthermore, the thesis proposes a Hausman type test to compare parametric with semiparametric estimators. The main results are as follows: all considered estimators delivered decent estimates of the average marginal effects, independent of the assumed functional form. The results for the estimation of marginal effects at specific points are different. The parametric estimators generally perform poorly, whereas the estimators derived from the true models perform well. Klein and Spady´s estimator performs decently in large samples. Moreover, a good performance with respect to the root mean squared error (RMSE) does generally not translate into a good estimation of the marginal effects.