A0505
Title: Shrinkage estimation for multivariate linear regression
Authors: Vali Asimit - City University London (United Kingdom) [presenting]
Abstract: Shrinkage estimation is a widely popular estimation procedure that is triggered by Stein's paradox, which had puzzled the statistical community for some time. The intuition behind is that different sources of information could be combined to "better" deal with a multidimensional estimation problem rather than separately estimate the individual estimation problems. We consider four shrinkage estimators to make "better" predictions for a multivariate linear regression model. Our simulation results and real data analyses show that our four shrinkage estimators are no worse than the classical Ordinary Least Square estimator, though at least one shrinkage estimator significantly improves the performance of the classical estimator. For example, the generalised linear model estimation massively benefits from our new estimators.
