Artificial intelligent based time series forecasting of stock prices using digital filters

The aim of the paper is the analysis of the sequential characteristics of the Athens Stock Exchange general index using the time series methodology based on artificial intelligent techniques. The applied models include the Feed Forward Neural Network trained with the efficient Levenberg–Marquardt optimization algorithm, the Adaptive Neuro-Fuzzy Inference System as well as traditional linear regression and ARIMA models for comparison. All these approaches are initially used for the short-term forecasting of the series, providing an insight into the forecasting capabilities of each model. The analysis of the spectral characteristics of the series indicated the presence of strong persistence or alternatively that the models do not differ significantly from a random walk. This observation was also cemented by the forecasting results of the developed models. The proposed approach is based on the application of low-pass digital filters on the series and the employment of the formerly mentioned models for the prediction of the created series. The filtered series contains a lower amount of noise and can be viewed as an alternative trend indication of the original series.