A0918
Title: Robustifying and simplifying high-dimensional regression: Applications to financial returns and telematics data
Authors: Michael Scholz - University of Klagenfurt (Austria) [presenting]
Jens Perch Nielsen - City, University of London (United Kingdom)
Malvina Marchese - City University of London (United Kingdom)
M. Dolores Martinez-Miranda - Universidad de Granada (Spain)
Abstract: The availability of a large number of variables that can have predictive power makes their selection in the regression context difficult. Robust and understandable low-dimensional estimators are considered as building blocks to improve the overall predictive power by combining these building blocks in an optimal way. The new algorithm is based on generalised cross-validation and builds the predictive model step-by-step forward from the simple mean to more complex predictive combinations. Practical applications to annual financial returns and actuarial telematics data show its usefulness for the financial and insurance industry.
