Multi-period portfolio optimization: A parallel NSGA-III algorithm with real-world constraints

Abstract

This study introduces an enhanced algorithm that integrates the parallel processing capabilities of PGAs with the multi-objective optimization strengths of NSGA-III, designed for multi-period optimization. We extend optimization objectives to T + 1 by minimizing risk over T periods and maximizing the terminal return, with a practical constraint on portfolio loss. It consistently outperforms the standard NSGA-III algorithm in both risk reduction and return optimization, especially when portfolios are adjusted quarterly. We also pinpoint optimal algorithmic parameters: a population size of 70 and 10 % migration rate. Overall, our research offers invaluable insights into real-world investment scenarios, serving both academic and industry interests.