The Chaotic Structure of Stock Markets and Forecasting with Neural Networks: The case of ISE-100

Abstract

In this study the aim is to determine whether there is a chaotic structure in the ISE-100 index return series for the period 23.10.1987-15.02.2011, and if it is determined that it exhibits chaotic behavior than the selection of the most successful prediction model. With this aim, first the existence of the chaotic behavior in the ISE-100 Index has been examined by several approaches and according to the results, it has been concluded that ISE-100 Index has chaotic characteristics. After, by using GARCH, EGARCH, feed-forward, recurrent and chaotic artificial neural network, forecasts have been performed for 5 and 15 days. The comparison of forecast successes has been made by various criterions and Diebold-Mariano Test. Findings proved that, the model of chaotic artificial neural network gives the most successful result for the ISE-100 Index.