中文
Published date:2014-03-27    Provided by:
 
Title: Second-order least squares estimation method for linear and nonlinear models
Guest SpeakerLiqun Wang, Department of Statistics, University of Manitoba
Time2013-12-27, 10:00-12:00
LocationRoom 204, Mechanics Engineering Building
Content &Introduction 

     In this lecture I will introduce the second-order least squares method for the estimation of regression models. While the ordinary least squares (OLS) method minimizes the quadratic distance of the response variable to its conditional mean given the predictor variables, the second-order least squares (SLS) method simultaneously minimizes the distances of the response variable and the squared response variable to its first and second conditional mean respectively. I will show that the SLS estimator is asymptotically more efficient than the OLS estimator when the third moment of the random error term is nonzero, and both estimators have the same asymptotic covariance matrix when the error distribution is symmetric. I will also present a simulation-based SLS estimator which is more practical and computationally feasible for models with complex structures.
     I will demonstrate that the SLS method will be extended to other linear and nonlinear models that are widely applied in applications, including censored regression model, errors-in-variables, mixed-effects models, and dynamic models.