Robustifying and simplifying high-dimensional regression with applications to yearly stock return and telematics data

Robustifying and simplifying high-dimensional regression with applications to yearly stock return and telematics data

Blog Article

Abstract 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 here blocks.Our new algorithm is based on generalized cross-validation and builds a predictive model step-by-step from a simple mean to more complex predictive combinations.

Empirical applications to annual financial returns and actuarial telematics data show its usefulness in the financial and lock shock and barrel art insurance industries.