A0933
Title: Beta regression: Shrinkage-Liu type estimator with application
Authors: Peter Karlsson - Linneaus University (Sweden) [presenting]
Stanislas Muhinyuza - Linnaeus University (Sweden)
Maziar Sahamkhadam - Linnaeus University (Sweden)
Abstract: Beta regression is widely used for modeling outcome variables bounded within the open interval from zero to one. However, when non-orthogonal regressors are present, the performance of the maximum likelihood estimator deteriorates, and a common solution to address this issue is to use a shrinkage estimator. Furthermore, sometimes, there is prior information about the parameters in the Beta regression model, such that the parameter vector is suspected to belong to a linear subspace. The restricted linear regression model with multicollinearity is common in practice, and shrinkage estimators of the model coefficients have been considered in the literature. However, the literature on shrinkage estimators for Beta regression models is limited under such settings. A two-parameter Liu linear shrinkage estimator is introduced, tailored for estimating the vector of parameters in a Beta regression model with a fixed dispersion parameter under the assumption of linear restrictions on the parameter vector. This estimator is particularly applicable in various practical scenarios where the level of correlation among the regressors varies, and the coefficient vector is suspected to belong to a linear subspace. Furthermore, the necessary and sufficient conditions for establishing the superiority of the new estimator over both one-parameter Liu estimators and two-parameter Stein-type estimators are derived. Finally, an empirical application is presented.
