Predicting Daily Stock Returns of Companies listed in Tehran Stock Exchange Using Artificial Neural Networks


This study has been conducted to investigate the predictability of stock returns behavior of the companies listed on "Tehran Stock Exchange" and also to predict the stock returns by using "Artificial Neural Networks". In order to predict the returns, in the first stage, the historical data relating to time series of the companies plus three variables of technical analysis (stock index, volume of exchanged stock and the closed price) for a five year period (July 1998 - 2003) were used. The optimum model for prediction of stock daily return for each company was designed via changes in parameters of the artificial neural network. In the second stage, prediction of daily return during the same period was done based only on past information or historical process of time series. In this study, "Multi Layered Perceptron(MLP)"artificial neural network with different learning functions were used.
The results showed that:
- The behavior of companies' stocks' daily return is not a random process and has memory.
- The artificial neural networks are able to predict the daily return with an acceptable error level.