Composition of investment portfolios through a combinatorial multiobjective optimization model using CVaR

TLDR

The study proposes a combinatorial multiobjective optimization framework for portfolio diversification that incorporates a downside risk measure. The authors implement parallel NSGA‑II and DEMO evolutionary algorithms on Ibovespa 2015 share portfolios, evaluating in‑sample diversity and objective‑space coverage and out‑of‑sample risk‑return trade‑offs across different portfolio sizes.

Abstract

This paper presents a combinatorial multiobjective optimization methodology to address diversification of investment in portfolios consistent with the market practices, using a "downside risk" measure. To cope with this feature, parallel versions of two evolutionary algorithms are proposed, based on NSGA-II and DEMO. Simulations consider portfolios comprised of shares that participated in the theoretical portfolio of Ibovespa in 2015. In-sample analysis considers graphical analysis, performance measures for diversity solutions and objective space coverage. Out-of-sample analysis is performed comparing the behavior of lower risk and higher return portfolios in relation to measures of risk and return, for several cardinalities.