Spin Glasses, the way to Distributed Processing Case Study on Stock Market Portfolio Selection

Subject Areas : Financial engineering

Majid  Vafaei Jahan ¹ (Associate Professor, Department of Computer Software, Faculty of Engineering and Engineering, Islamic Azad University, Mashhad Branch)
mohammadreza  Akbarzadeh Tutunchi ² (Professor, Department of Electrical Engineering, Faculty of Engineering and Engineering, Ferdowsi University of Mashhad)

Received: 2017-02-15 Accepted : 2017-04-29 Published : 2017-09-23

Keywords: Learning Automata, Simulated Annealing, Portfolio Selection, Spin glass Model, Extremal Optimization,

Abstract :

The several heuristic algorithms have been proposed for portfolio selection. One of these algorithms is based on spin glasses that have local searching and parallel processing properties. Because of the spin glass algorithms are actually based on Monte Carlo simulation such as simulated annealing (SA) and have low convergence speed against other method, yet composing with other methods such as Learning Automata (LA) and genetic algorithms have been considered. In this paper, one of the composing methods based on SA and Exteremal Optimization (EO) has been proposed, this algorithm select and change the low order spins with higher probability and take the state of all spins into the better situation. After a sufficient number of steps, the system reaches a highly correlated that almost all species have reached fitness above a certain threshold. This co-evolutionary activity gives rise to chain reactions and every fluctuation that rearrange major parts of the system, potentially making any configuration accessible. Therefore any fluctuations allow escaping from local minima and efficiently explore the configuration space. The experimental results show this method is powerful paradigm for finding ground state of spin glass and better than other methods such as SA and LA for solving portfolio selection problem.