comparison of the ability of modern and conventional metaheuristic and regression models to predict stock returns by accounting variables and presenting an effective model

Investment in the stock market requires decision-making and access to information on the future of the stock market. given the importance of predicting stock returns, the present study aimed to discover the variables and indices that could predict stock returns. the prediction of stock returns has long been a 'hot topic' in development countries. while effective steps have been taken in this regard, the accurate prediction of stock returns remains a problem due to numerous issues. in this study, an accurate, applicable, and effective model was proposed for the prediction of stock returns. the statistical sample included 138 active companies listed in tehran stock exchange (tse) during 2008-2017. in total, 1,380 data years were selected for the research to evaluate the questions. data analysis was performed using an adaptive neuro-fuzzy inference system (anfis), multi-gene genetic programming, and regression analysis. in addition, statistical tests were applied to evaluate the accuracy of the model, implemented by matlab and gene x pro tools. according to the results, the hybrid metaheuristic method had a lower error rate compared to artificial neural network and regression analysis in terms of stock return prediction. therefore, the proposed model could provide more accurate data within a shorter time to predict the stock market status since it makes predictions after selecting the most optimal input variables through anfis.