| Summary: | 碩士 === 輔仁大學 === 資訊管理學系 === 96 === In previous research, Particle Swarm Optimization (PSO) was conceptually simpler than Genetic Algorithms (GA) and can use lesser parameter in model design. The PSO usually performs well in the early iteration, but the disadvantage of PSO was that it was unable to dynamicly adjust the speed of movement of partical when partical reached a near optimal solution in several function optimization problems. Besides, another problem is that partical swarm was easily premature convergence in multimodal optimization problems. Therefore, the aim of this research combined the strengths of PSO with GA. The hybrid algorithm, PSOGA, combined the standard velocity and position update rules of PSOs with the ideas of selection, crossover and mutation from GA.We proposed a new velocity update rule, PSOGA_Lbest, incorporating with the impact factor of concept to improve swarm learning mechanism in standard PSO. This research used the PSOGA and PSOGA_Lbest to design and construct the “best portfolio of Mutual Funds”, in order to inspecting the new algorithms evolution mechanism.We hoped this system can design the best portfolio with high rate of returns and relatively low risk simultaneously for investor.
The experimental result showed that performance of profolio of mutual funds (high rate of returns with low risk) improved PSO (PSOGA and PSOGA_Lbest) designed was better than the domestic and offshore fund of funds and greater then TSEC weighted price index. Then, the improved PSO was highly competitive and stable, often outperforming tradition PSO and GA. Therefore, professional funds manager engaged in designing and operating mutaul funds merchandise can use this model to make decisions or poduce suggestions.For investors with self-investment attribute, they can deploy portfolio of mutual funds through this system.
In addition, we found that the convergence efficiency of PSOGA is better in performance through observing the evoluntionary characteristic in this research .Then, PSOGA_Lbest let the particle scatter more widely in solution space. And reached better optimization in multi-objective explanation space , on the other hand , PSOGA was more effictiveness and efficiency.
|
|---|
