Neurodynamics-driven portfolio optimization with targeted performance criteria.

Abstract

This paper addresses portfolio selection with targeted performance criteria via neurodynamic optimization. Five portfolio optimization problems are formulated with a variable weight to maximize five risk-adjusted performance criteria in Markowitz's mean-variance framework and reformulated as iteratively weighted convex optimization problems to facilitate subsequent problem-solving solution procedures. In addition, distributed portfolio optimization problems with separable performance criteria are also formulated. Three neurodynamic approaches are developed based on two globally convergent recurrent neural networks to solve the formulated and reformulated problems. Extensive experimental results on 13 datasets of world stock markets are elaborated to demonstrate the superior performance of the neurodynamic approaches against the baselines in terms of five given evaluation criteria and two investment returns.