Comparative Evaluation of Markowitz Approach with a New Hybrid Method to Create an Optimal Portfolio Using Deep DNN Learning Method and Gravitational Search Algorithm

The aim of this study is to compare the New Hybrid Method with the usual Markowitz method in creating an optimal portfolio. To this end, at the first stage, the future stock prices were predicted using a deep DNN learning method and stock technical variables for the period 1397/4/2 to 1397/6/2. Then, based on future stock prices, stock return and risk were calculated and, by using Gravitational Algorithm, portfolio profits were maximized. This results in creating low risk to high risk portfolios on the Pareto efficient frontier. After that, the future return of portfolios was calculated for the next two months, and the process was repeated for 30 weeks in the form of weekly Rolling Window. These results were compared with the results of usual Markovitz method for 30 periods. The results indicated that both Markowitz and New Hybrid methods showed only better performance in predicting stock prices of risk averse portfolios than average market index.