Time Series Forecasting of the Austrian Traded Index (ATX) Using Artificial Neural Network Model

This paper analyses the Austrian Traded Index (ATX) of the Vienna Stock Exchange (Wiener Börse) in the period from 2009 to 2017, using the method of the artificial neural network (ANN). Sampling data are taken from the web page of the Wiener Börse and filtered on weekly basis to comply with weekly seasonality in eight years range. The aim is to construct several AAN models that meet certain criteria and evaluate them on the holdout subsample. Furthermore, the goal is to find the best model that can predict new upcoming yet unseen data with high accuracy. A data frame for testing forecasting performance is one month, a quartile, a half year, and one year period for which last year of the data sample is retained (August, 2016- August 2017). Using various criteria and different parameters, the total of thirty networks were built and tested and top five networks were analysed in more details. Results confirm high accuracy of using method of artificial neural networks, which is consistent to studies conducted on similar cases. Correlation of top three selected networks by validation subsample is over 0,9. The mean absolute percentage errors (MAPE) for the best selected network are 1,76% (month); 2,11% (quartile); 2,21% (half-year); 2,13% (year). Once again, ANN method has proven to be a powerful forecasting tool