Applying machine learning models in creation of share optimum portfolio and their comparison

Although econometric models are appropriate for describing and evaluating the relationships between variables and statistical inference, but they have some limitations for financial analysis. Many efforts have been made to model nonlinear relationships in financial data using machine learning technologies. The purpose of this study is to apply machine learning models to form optimal stock portfolios and compare their performance. The statistical sample of the present study consists of 156 companies listed in Tehran Stock Exchange during the period 2009-2018. After data collection, the intended deep learning models in Anaconda software and Python programming language were tested, and then the ability of each model was determined by return evaluation, composite return, trenors and jensens criteria to form an optimal stock portfolio. According to the free-risk and market return rate, forming portfolio by investor to more profit than these two rates and portfolio valuation results of trenors and jensens indexes, it was concluded that the deep Convolutional Neural Network is able to for optimal portfolio. According to this reasoning, the long short-term memory model is not capable of optimal portfolio formation.