Title: Sorting out your investments: Sparse portfolio construction via the ordered L1-norm
Authors: Philipp Johannes Kremer - EBS Universitaet fuer Wirtschaft und Recht (Germany) [presenting]
Sandra Paterlini - European Business School Germany (Germany)
Malgorzata Bogdan - Lund University (Sweden)
Sangkyun Lee - TU Dortmund University (Germany)
Abstract: Since its introduction to the statistics literature, the desiring features of simultaneous model selection and estimation have gained Lasso a wide recognition in statistics and also recently in financial portfolio optimization. Still, the Lasso has well-known shortcomings when applied to the setting of highly dependent financial data. We move away from the standard framework of orthogonal design and apply the recently developed sorted $l_1$ penalized estimation, called SLOPE, to the framework of correlated data. SLOPE relies on the idea of penalizing coefficients with a stronger signal more heavily and clumping equally correlated assets together. In fact, in a simulated factor model, SLOPE is able to identify and to cluster assets with the same underlying risk factor exposures into one group. This enables the investor to improve his ex-ante portfolio risk management. Furthermore, our empirical analysis on the SP100 and SP500 from 2004-2016 confirms the validity of SLOPE in developing effective investment strategies.