Improved Set-based Particle Swarm optimization for Portfolio optimization

Recently, a set-based particle swarm optimization algorithm was proposed for portfolio optimization. The algorithm used a bi-stage search process that first selected assets and then determined the weights for said assets. Unlike other approaches before it, weight determination was accomplished by the use of a nature-inspired algorithm as opposed to quadratic programming. The algorithm was shown to be capable of obtaining goodquality solutions while being relatively fast. However, there were several aspects of the algorithm that could be improved upon, namely, convergence, efficiency, and in some cases, the ability to approximate the known Pareto optimal front. This paper suggests and investigates two modified versions of the set-based algorithm that are shown to yield substantial gains in performance. The latter, referred to as improved set-based particle swarm optimization, was able to find portfolios that are more, or as profitable as those obtained by the original approach, but faster and with lower levels of risk.