A Hybrid Model of Stochastic Dynamic Programming and Genetic Algorithm for Multistage Portfolio Optimization with GlueVaR Risk Measurement

Abstract

Objective: The selection of an optimal investment portfolio for a long-term period does not seem logical. So the investors should update their investment portfolios over specific time periods if needed. Since the problem dimensions significantly increase after the periods, a definitive solution to the problem is not achievable. Methods: In this regard, the Multistage Approximate Stochastic Dynamic Programming has been used to make the best portfolio over each period by using a stochastic return rate. The Monte Carlo was used for scenario development, and GlueVar was selected as a risk measurement criterion. The approximation technique was used to resolve for large dimensions; however, some optimized solutions may be eliminated so we used the Genetic Algorithm for the rapid search around the optimal solution to obtain a better one, if possible. Results: Top 100 companies listed in the Tehran Stock Exchange between 2011 and 2017 were investigated. This study investigated and compared the return and risk of investment portfolios based on the proposed method, Genetic Algorithm, and stock portfolio with equal weights. The modeling was done with MATLAB and tests were carried out with SPSS. Conclusion: The results indicated a higher performance of the proposed method in comparison with the other mentioned methods.

Keywords

• portfolio optimization
• stochastic dynamic programming
• gluevar risk measurement
• genetic algorithm
• scenario constructiom