Abstract: Stock price developments are a nonlinear and dynamic process. Stock price prediction is one of the most important issues in stock markets. The aim of this paper iis paper aims to verify the prediction of the stock prices of ČEZ, a. s. on the Prague Stock Exchange. The main goal is decomposed into two sub-goals: evaluate the accuracy of the prediction using neural networks on stock price data that followed the time series used to calculate neural networks in the contribution of Vrbka and Rowland (2017) and generate new neural structures. 1,442 records of price data on stock are used. Multilayer Perceptron Neural Networks and Neural Networks of Basic Radial Functions are generated. The original neural networks are inappropriate to predict the evolution of ČEZ stock prices. Optically, two new neural structures can be used in practice. The practical use of new networks is not tested on data that did not enter the calculation. New networks show lower performance in all three sets of data.
Authors: Petr Šuleř, Jakub Horák, Tomáš Krulický
Keywords: Artificial neural networks, prediction, stock prices, time series, Czech Republic