OPTIMUM PORTFOLIO CONSTRUCTION CONSIDERING VALUE AT RISK USING GENETIC ALGORITHMS AND MONTE CARLO SIMULATION

Determining the best portfolio out of set of alternative investment opportunities to optimize risk-adjusted return and value-at-risk simultaneously is a challenging issue for many practitioners. In recent years, the application of non-conventional methods for portfolio optimization problems has grown in importance in the investment industry. As an effective alternative to traditional optimization techniques for handling the computationally complicated portfolio optimization problems, many nature-inspired optimization methods have emerged and have been developed by researchers. In this thesis, a novel algorithm is suggested to construct a promising portfolio in terms of Mean return- VaR and Sharpe ratio-VaR from a limited number of securities from a set of available equities. The algorithm consist of three stages of refining. The first stage is to select 60 stocks out of all the securities in S&P500 index based on fundamental factors using factor analysis. In the second stage, the proposed algorithm employs a Markowitz' mean-variance optimization model to refine the quality of initial population of portfolios of 30 stocks and improve the convergence behaviour of the algorithm. And in the third stage, a state-of-the-art genetic algorithm is applied to determine an optimized portfolio of assets in terms of risk-adjusted return and value at risk. The novel genetic algorithm developed in this research benefits from an innovative solution representation which make GA searches over both discrete and continuous variables in the problem of optimizing stock and industry selection and weight allocation. In this study, the outperformance and effectiveness of the proposed algorithm are demonstrated by comparing annual return, annual volatility, Sharpe ratio, Jensen's alpha and beta of a constructed portfolio with the S&P 500 index and Mean-Variance constructed portfolio. The robustness of our evolutionary algorithm is verified by evaluation of the results in both in-sample and out-sample data.