Portfolio optimization strategy based on improved particle swarm optimization algorithm and SA fusion
Abstract
Traditional portfolio optimization models have limitations in addressing high-dimensional non-convex issues and complex real-world constraints, such as parameter sensitivity, susceptibility to local optima, and strong assumptions. To achieve more robust and efficient asset allocation, this paper constructs a hybrid optimization model that integrates improved particle swarm optimization (PSO) and simulated annealing (SA) algorithms. This method enhances population diversity by introducing complex methods, balancing exploration and development with adaptive inertia weights and annealing strategies, and embedding metropolis criteria to probabilistically accept inferior solutions, thereby effectively improving global convergence performance and solution stability. The experiment showed that the SA–PSO model significantly outperformed the compared algorithms in multiple performance dimensions. In the Sharpe ratio and Sortino ratio tests, the investment success rates reached 94.5% and 96.2%, maintaining a stable investment rate of over 90% in different market environments, and the response time was shortened by more than 40% compared to traditional PSO and genetic algorithms. In addition, even with the addition of complex constraints, the model still maintained over 90% investment accuracy, significantly outperforming commercial software such as Matrix Lab and Python commercial solvers. The conclusion was that the SA–PSO model effectively overcame the shortcomings of traditional optimization methods and providing an effective solution framework under the considered conditions for dealing with multi-constraint and dynamic market problems in actual investment. This study has important practical significance in promoting the application of intelligent optimization theory in asset management, and provides reliable methodological support for building investment decision-making systems with strong adaptability and high robustness.
