The availability of many variables with predictive power makes their selection in a regression context difficult. This study considers robust and understandable low-dimensional estimators as building blocks to improve overall predictive power by optimally combining these building blocks. Our new...