Examining the efficiency of optimization models of multi objective genetic algorithm and particle swarm algorithm under the risk criteria of conditional value at risk and mean smai variance in determining the optimal stock portfolio

The goal is to select an optimal portfolio of stocks by allocating capital among various investment opportunities in the stock market to achieve maximum return at a specified level of risk. This constitutes an efficient portfolio.

Attaining an efficient portfolio involves solving an optimization problem. There are numerous techniques and tools available to solve this issue. In this study, 15 stocks from companies listed on the Tehran Stock Exchange, including symbols such as Khapars, Khazamiya, Vepasar, Foulad, Akhabar, Kegel, Femli, Tapiko, Sepaha, Fazer, Fakhas, Shohbaran, Shefan, Qamro and Qathabat, were selected using cluster sampling. First, the daily returns of these stocks were calculated over a 5-year period from 2015 to 2020 (1183 days). The risk of the optimal investment portfolio was then calculated using the Mean-Semi Variance and Conditional Value at Risk models. These two criteria were compared using a classic solution method. Subsequently, the output data obtained from these calculations were compared using MATLAB software, employing the Particle Swarm Optimization algorithm under the Mean-Semi Variance risk criterion and the Genetic Algorithm under the Conditional Value at Risk criterion.

The results of this study indicate that the meta-heuristic Particle Swarm Optimization method yields a higher portfolio return ratio compared to the Genetic Algorithm in the Mean-Semi Variance risk criterion.

Originality / Value: 

This research utilizes multi-objective genetic algorithms and Particle Swarm Optimization, which are intelligent and novel algorithms, to minimize the objective function value using Conditional Value at Risk and Mean-Semi Variance criteria. These algorithms optimize the return and risk ratios of the stocks in the investment portfolio with the highest possible accuracy. Additionally, the efficiency comparison of these models using MATLAB software contributes an innovative aspect to this study.