Developing a neural network system for financial prediction and an application on ISE

Abstract

After a number of successful applications in image processmg and recognition, neural networks have gained a wide-spread use and an admirable place in data processing field. Their ability in data mining, classification, generalisation and trend prediction is the key factor of which they have been extensively used in finance. As neural networks are good at learning non-linear relationships and at predicting non-random movements, they have been utilised especially in forecasting stock price movements. Even though most of the researches and the models gave unsatisfactory results, there are a few successful applications that encourage the use of neural networks in the prediction of stock price movements. In this study, it's explained how artificial neural networks are used as a method for financial forecasting. After giving a brief description of neural networks, every step of developing a neural network forecasting system is explained in detail. In this study, four models have been developed for the prediction of stock price movements, using multi-layer feed-forward neural networks. By presenting the long and short term trends of the end-of-week closing prices, weekly trade volume and the end-of-week market index to the network, the models are expected to predict the future price movements by analysing the past trends. At last, a model has been developed for the prediction of market index.