Inspecting the Predictive Power of Artificial Intelligence Models in Predicting the Stock Price Trend in Tehran Stock Exchange

Time series prediction methods based on artificial intelligence have been widely developed in recent years. Given that these data have large dimensions in the field of investment and stock price forecasting, traditional data analysis methods have low predictive power. This study examines the predictive power of a variety of models based on machine learning in the Tehran Stock Exchange.

After collecting data from 150 large companies listed on the Tehran Stock Exchange from 2012 to 2021, we want to predict the stock price trend- the movement direction of the price- and then validate each method and compare their accuracy. In these methods, we allocate part of the data to the learning section and the rest to the test section. We take these periods as training and trading sets. These methods include linear models, autocorrelation models, trees, and neural networks.

Deep learning models show better performance than other models and have an accuracy of about 70 percent. Also, we show the time series of the best-performance model accuracy of portfolios of some large industries. The best-performance model of DL in this study is Recurrent Neural Networks. In addition, we show that shallow learning models have higher accuracy and most models perform better in predicting descending stock trends.

In this study, after trying to use the models very carefully, the result is that these models do not provide stunning results to investors.