infonet-economy

Not available in German. The paper describes the adaptation of the estimation method proposed by Gallant and Nychka (1987) for the selectivity model to the threshold crossing model of binary choice under independence between errors and regressors. We present Monte-Carlo and asymptotic comparisons with the probit estimator and discuss appropriate maximization algorithms, suitable choice of starting values and strategies for the choice of the number of parameters used in approximating the density. Semi-nonparametric estimation is almost as efficient as probit estimation in normal samples and performs much better in non-normal samples. We also use the method for a participation model estimated on 3658 observations with 21 explanatory variables and show that it is practicable on modern personal computers. Pseudo score test results based on this methodology are presented, with special attention to heteroscedasticity as the main remaining potential cause for inconsistency.