Forecasting OMX Vilnius stock index – a neural network approach

Abstract

Predicting a stock market is a challenging task for every investor. Stock market contains difficult relations and its behavior is heavily forecasted. As the investment’s profitability is directly related to the market’s predictability, the need for more accurate and sophisticated forecasting techniques arises. The academic literature is showing a growing interest in implementing non- linear techniques in a time series prediction. The paper goes through the process of creating a time series prediction model for OMX Vilnius stock index using artificial neural network approach. A multi layer perceptron model is applied in order to make periodical daily and monthly forecasts for both the actual index future value and the direction of the index. The neural network is trained using back-propagation method, several topologies are analyzed and the most suitable is selected. The method accuracy is compared to several traditional statistical methods (moving averages and linear regression).