Collaborative Neurodynamic Approaches to Portfolio Optimization

Description

Portfolio selection is of great interest for financial investments from academic and economic points of view. Over the past six decades, portfolio selection has been extensively explored, highlighted by the works of at least five Nobel laureates in economics. As a cornerstone of finance, modern portfolio theory featured by the Markowitz’s ground-breakring work on mean-variance analysis provides a formal framework for portfolio optimization with quantified investment returns and risks. Despite its theoretical importance, the mean-variance theory is not perfect, as some of its assumptions are unrealistic. Because investors’ risk behaviors toward gains and losses are usually asymmetric and underlying probability distributions are non-normal, variance is not an ideal risk measure to capture risk behaviors toward gains/losses. Although several alternatives are comparably reasonable, the nonconvexity and discontinuity of some measures post great challenges for portfolio optimization mathematically and computationally. In parallel to research on portfolio selection in finance, research on artificial intelligence in science and engineering has been active for decades and becomes very popular in recent few years due to the successes of deep learning. As a stream of neural network research, neurodynamic optimization has grown as a parallel and distributed optimization approach and many recurrent neural networks have been developed for solving various optimization problems. Our recent research breakthroughs show that neurodynamic systems composed of multiple neural networks coordinated with rules in swarm intelligence are efficacious for constrained global, multi-period, and multiple-objective optimization. The results are being extended to bi-level and combinatorial optmization. The integration of neurodynamic optimization and portfolio management would lead to fruitful results and important breakthroughs. Despite the progresses in neurodynamic optimization, neurodynamics-based portfolio optimization deserves in-depth investigations in its own right because of the distinctive complexities in depth and scale of financial engineering and management. In this proposed research, based on neural computation, systematic investigations will be performed on intelligent methodology for portfolio selection and related problems. The project will consist of five coherent parts focusing on the research and development of collaborative neurodynamic approaches to portfolio optimization in five major settings. The expected outcomes of the proposed research include new insights, convergent models, innovative methods, and integrated systems for portfolio optimization. The conceived neurodynamics-based intelligent systems for portfolio optimization would serve as viable problem-solving tools for institutional and individual investors. It is envisioned that the successful completion of the proposed project would significantly advance the frontiers of portfolio optimization in both theory and practice.