Mixtures of Regression with Unknown Symmetric Error Density

• Activity Number: 459
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Weixin Yao*+ and Shaoli Wang and Chunrong Ai
• Companies: Kansas State University and Shanghai University of Finance and Economics and University of Florida
• Keywords: Mixtures of regression ; Mixture models ; EM algorithm ; Semiparametric model

Abstract

Mixtures of regression are widely used to investigate the relationship between interested variables if they are from several unknown latent groups/clusters. The existing estimation procedures for the mixtures of linear regression depend on the normality assumption of the error density of each component. In this talk, we will propose a semiparametric mixture of regression model which only assumes that the error density within each component is symmetric. A semiparametric EM algorithm is proposed to estimate the mixtures of linear regression parameters. An extension of our proposed model and estimation procedure to the mixtures of nonparametric regression is also considered. From the the simulation results, we will see that our proposed procedure works reasonably well when the error is normal and works much better than the existing estimation method when the error is not normal.