Predicting prices of stocks listed in Tehran Stock Exchange using Generative Adversarial Networks
Subject Areas : Financial Knowledge of Securities Analysisرضا راعی 1 , ali NAMAKI 2 , Saeed Bajalan 3 , sara najafe zade 4
1 - Professor of Finance, Finance and Insurance Department, Faculty of Management, University of Tehran, Tehran, Iran. (Corresponding Author)
2 - Assistant professor of Finance, Finance and Insurance Department, Faculty of Management, University of Tehran, Tehran, Iran.
3 - Assistant professor of Finance, Finance and Insurance Department, Faculty of Management, University of Tehran, Tehran, Iran
4 - Master of Finance, Finance and Insurance Department, Faculty of Management, University of Tehran, Tehran, Iran
Keywords: Deep learning, generative adversarial network, stock market prediction, financial markets.,
Abstract :
Investors are willing to invest in the capital market which would earn a proper profit and would make the possibility of an accurate forecast of future trends and prices. In this regard, deep learning (DL) networks have been able to help in predicting the capital market movement. Due to the high capacity of DL approaches in many fields due to their strong capacity, they have been widely used in financial issues such as stock market movement prediction, portfolio optimization, financial information processing, etc. Recently, generative adversarial networks (GANs) illustrated suitable results intending to analyze and predict time series data. Therefore, in this study, the GAN consisting of a convolutional neural network as a generator and long short-term memory in the adversarial network is proposed for stock price prediction. In addition, in order to increase the accuracy of the network, other DL approaches have been used in network training. The results of the daily data of the Tehran Stock Exchange between 1394 to 1398 demonstrate that the prediction accuracy of the GAN network using the most appropriate features is up to 10%. The experimental results of this model show that the GAN network with the mentioned architecture can have a promising performance in stock price prediction compared to other DL models.
