A Time Step Cascade CNN-LSTM neural network for predicting adjusted close price of 5 largest firms in Tehran stock exchange

Tehran stock exchange has gained a lot of attention through recent years. This is because of the commercial benefits that it has for individual investors and investment firms. Although many Artificial Intelligence and econometrics researchers have already published various articles with the aim of market prediction, there are many uncovered approaches left. In this research a deep multivariate time step CNN-LSTM model is introduced to study historical patterns of market data. The constructed model is applied on 5 of the largest firms in Tehran stock market, and the study period spans from 1 January 2010 up to 1 December of 2020. The main aim of this paper is to find out the ticker that can be predicted accurately by the introduced model. Results from MAE, MSE, MAPE, R2 score and residual diagnostic test has shown that Ghadir Inv can be predicted better than other tickers by the introduced model even through the recent financial crisis of Tehran stock exchange.

Keywords:

CNN-LSTM , Tehran stock exchange , time series prediction , deep learning