Robust Portfolio Optimisation Under Sparse Contamination

We introduce novel methods for mean-variance portfolio optimisation in the presence of component-wise contamination. Methods are obtained by combining component-wise robust location-scatter estimation and optimisation based on genetic algorithms. The newly proposed approaches are compared with classical and row-wise robust methods in a simulation study and a real-data application on data from the Italian stock exchange. Results show a strong advantage of cell-wise resistant methodologies over competitors, both in terms of absolute risk and Sharpe ratio.