Hybrid robust portfolio selection model using machine learning-based preselection

Robust portfolio optimization theory is an es-sential foundation for modern financial modeling, which is a well-studied but not fully conquered territory. Conservatism is one of the most discussed issues by numerous scholars. To obtain a robust portfolio model with satisfactory perfor-mance, we propose the hybrid robust mean-variance portfolio model constrained with different ellipsoidal uncertainty sets in this paper. Additionally, skewness is also considered in the objective function. Preselection is designed for picking out the high-quality risky assets, where two machine learning algorithms, Random Forest and Support Vector Machine, are involved. In the numerical experiments, the US 48 industry data set from Kenneth R. French is employed to verify the effectiveness of the proposed hybrid portfolio models. The comparative results between the proposed hybrid models and baseline portfolio models (equal-weighted model, mean-variance model, mean-variance-skewness model) show that the proposed hybrid robust mean-variance portfolios considering skewness with preselection beat the baseline strategies by a clear margin. Also, the actual effectiveness of skewness in the hybrid robust models is analyzed. © 2021, International Association of Engineers. All rights reserved.