On Mixture Regression Shrinkage and Selection Via the MR-Lasso

Abstract

In finite mixture regression models, we generalize the application of the least absolute shrinkage and selection operator (LASSO) to obtain MR-Lasso, which incorporates both mixture and regression penalties. Because MR-Lasso jointly penalizes both regression coeficients and mixture components, it enables simultaneous identification of significant variables and determination of important mixture components. Simulation studies indicate that MR-Lasso outperforms LASSO. Extensions to mixture non-Gaussian and mixture time series models are briefly described.

Keywords: finite mixture model, LASSO, mixture penalty

JEL Classification: C5, C52