Predictability of Tehran Stock Exchange using deep learning models (CNN-LSTM model)

Deep learning is a subset of the broader class of neural network based machine learning methods that has recently gained much attention in many fields, including time series prediction in financial markets. In this research, first, using deep learning models based on LSTM and CNN networks, the movement of Tehran Stock Exchange index is predicted. Then, by combining the two models, the hybrid CNN-LSTM deep learning model is used to predict the Tehran Stock Exchange index. In the next step, in order to evaluate the performance of the aforementioned forecasting models, three performance measures symmetric mean absolute error percentage (SMAPE), mean absolute error percentage (MAPE) and root mean square error (RMSE) were used. In this research, the daily data of the Tehran Stock Exchange Index was used in the period of 2016-07-13 until 2021-01-26. The estimation results of the models in predicting the Tehran Stock Exchange index with a one day step and comparing the efficiency measurement criteria indicate the superiority of the proposed CNN-LSTM model compared to the other two models. The LSTM model ranks next in accuracy and forecasting efficiency. According to the results that be presented in this research, financial market participants in Iran are suggested to pay attention to integrated deep learning models in order to increase the efficiency and accuracy of their predictions.