Mofid University The Journal of Economic Studies and Policies 2423-4648 0 9 2006 06 22 A Comparative Study of Neural Networks' Capabilities Using Indicators of Technical Analysis for Forecasting of Stock's Price A Comparative Study of Neural Networks' Capabilities Using Indicators of Technical Analysis for Forecasting of Stock's Price 57 82 47060 FA Mahmood Motevasseli Professor of Economics, University of Tehran Bijan Taleb Kashefi Master of Economics Journal Article 2006 04 04 The most important issue for active investors in capital market is forecasting the stock's price. The main goal of this research is to study the application of stock's price anticipation by using indicators of technical analysis based on neural networks and the comparison of this method and neural networks which uses stock's price and ARIMA models. In this research stock's price of the next 10 days of 40 active companies will be anticipated in Tehran stock exchange by using three different methods. In the first method, the stock's price will be forecasted by applying SINGLE LAYER FEED FORWARD NEURAL NETWORKS. Using Levenberg-Marquardt learning algorithm and the performance criteria of   MSE with admission of market value. <br />In the next step, beside the entry of market value, 5, 10 and 20 days of moving average and 12 days of RSI and also ROC were introduced as new entries to the network and forecasting was accomplished. The Stock's prices were also anticipated for all companies using   ARIMA models. By   applying analysis of variance, three different anticipating models were compared. Since price anticipation for thirty companies by ARIMA models have presented better and more meaningful results rather than neural networks model. Therefore, we can claim that the linear models of ARIMA are more capable of explaining and analyzing the complexities of time series of stock's price than the nonlinear models of neural networks. Therefore, they are recommended to be used for anticipation of stock's price The most important issue for active investors in capital market is forecasting the stock's price. The main goal of this research is to study the application of stock's price anticipation by using indicators of technical analysis based on neural networks and the comparison of this method and neural networks which uses stock's price and ARIMA models. In this research stock's price of the next 10 days of 40 active companies will be anticipated in Tehran stock exchange by using three different methods. In the first method, the stock's price will be forecasted by applying SINGLE LAYER FEED FORWARD NEURAL NETWORKS. Using Levenberg-Marquardt learning algorithm and the performance criteria of   MSE with admission of market value. <br />In the next step, beside the entry of market value, 5, 10 and 20 days of moving average and 12 days of RSI and also ROC were introduced as new entries to the network and forecasting was accomplished. The Stock's prices were also anticipated for all companies using   ARIMA models. By   applying analysis of variance, three different anticipating models were compared. Since price anticipation for thirty companies by ARIMA models have presented better and more meaningful results rather than neural networks model. Therefore, we can claim that the linear models of ARIMA are more capable of explaining and analyzing the complexities of time series of stock's price than the nonlinear models of neural networks. Therefore, they are recommended to be used for anticipation of stock's price https://economic.mofidu.ac.ir/article_47060_4140fe6a9358de3c9d660225c4c2fcb4.pdf