9321103 Wang Existing semi-parametric methods for binary choice models face two major disadvantages: namely, computational difficulty and multiple local maxima, especially for models with multiple explanatory variables or large sample sizes. Another problem is that some existing semi-parametric methods need a very large size to obtain the benefits promised by asymptotic theory. These problems are so severe that there is hardly any empirical application of semi-parametric methods for binary choice models available in the literature although these methods have been developed for a while. This project proposes a computationally convenient semi-parametric estimation method for binary choice models based on a semi-parametric interpretation of the EM (Expectation and Maximization) principle and the least squares approach. Preliminary Monte Carlo studies show that this new method has no problem of multiple local maxima (especially for large sample sizes) and is computationally much less time consuming than other estimators. This project studies the theoretical properties of the new estimator of binary choice models, including large sample properties and the use of resampling methods for estimating the variance of the estimator. The use of optimal weighted least squares estimators instead of Ordinary Least Squares (OLS) will be investigated as a means of improving efficiency. Extensive simulations will be conducted in order to compare the performance of various semi-parametric estimators. This method will be applied to an important empirical economic problem, the early retirement decision. To facilitate the application of the proposed method, the project will develop efficient and convenient computer programs for the new estimator. These programs will be easily incorporated in standard econometric computer packages, such as SAS, GAUSS and LIMDEP.
