Neurodynamics-driven portfolio optimization with targeted performance criteria.

Area of Science:

• Computational Finance
• Artificial Intelligence
• Optimization Theory

Background:

• Traditional portfolio optimization faces challenges with complex performance criteria.
• Markowitz's mean-variance framework is a cornerstone but requires efficient solution methods for advanced applications.
• Neurodynamic approaches offer a novel computational paradigm for financial optimization.

Purpose of the Study:

• To develop and evaluate neurodynamic optimization techniques for portfolio selection with targeted performance criteria.
• To reformulate portfolio optimization problems into solvable convex optimization formats.
• To address distributed portfolio optimization with separable performance criteria.

Main Methods:

• Formulation of five portfolio optimization problems with variable weights for risk-adjusted criteria.
• Reformulation into iteratively weighted convex optimization problems.
• Development of three neurodynamic approaches using two globally convergent recurrent neural networks.

Main Results:

• Neurodynamic approaches demonstrated superior performance compared to baseline methods.
• Effectiveness validated across 13 global stock market datasets.
• Improvements noted in five evaluation criteria and two investment return metrics.

Conclusions:

• Neurodynamic optimization is a powerful tool for complex portfolio selection tasks.
• The proposed methods offer an efficient and effective alternative to existing techniques.
• This research advances the application of artificial intelligence in financial optimization.