Usage of Artifical Intelligence and Spectral Analysis for Predicting the Behavior of Stock Prices

Abstract

In this paper methods of artificial intelligence and spectral analysis to build an algorithm for predicting the behavior of stock prices are applied. Spectral decomposition of a time series was calculated using known methods based on Fourier transformation. The results obtained from periodogram analysis simply provide information about periodicities. Significance analysis was not performed and we worked with four frequencies. This spectral information is then used in clustering of data. Comparison of behavior of price oscillation in clusters was carried out. The presented contribution aims to describe a new algorithm for predicting the behavior of stock prices. The clustering algorithm is based on spectral analysis and SOM. The whole procedure is tested on selected time sections of Dow Jones Industrial Averages, where the algorithm is performed. Results of analysis and final discussion, presented in the Case Study, show that the new method successfully signalizes the trend of stock market prices.