In this talk, we tackle a sustainable multi-objective optimization problem in which we aim to maximize a mean-risk ratio while, at the same time, minimizing the environmental impact.
More specifically, the mean-risk metric is represented by the Stable Tail Adjusted Return ratio, which penalizes only downside tail-risk and incorporates skewness and fat-tail information. In the portfolio construction, we consider the classical constraints concerning full investment and buy-in threshold requirements.
To deal with these new asset allocation models, we develop an improved Multi-Objective Particle Swarm Optimizer (MOPSO) embedded with a repair-projection mechanism to satisfy the constraints. Moreover, we implement a deep learning architecture to improve the estimation quality of the mean-risk measure and to speed up the MOPSO solver.
Finally, we conduct empirical tests on a European dataset to illustrate the effectiveness of the proposed strategies, accounting for different levels of environmental awareness.