ANALYSIS OF THE STOCK MARKETS STRUCTURE USING NEURAL NETWORKS

<p style="text-align: justify;">The paper presents the solution of the data classification problem for U.S. stock market and the BRIC countries stock markets, based on neural network approach. We considered stock markets data for the period 2007-2011. A multilayer feed forward neural network with RPROPS algorithm is introduced for the data classification. The input values for the neural network were presented in the vector form. Each vector element characterizes the number of stocks owned by the respective interval of the correlation distribution histogram. The outputs of the neural network were represented by countries stock markets for the vectors. The time scale was divided into time overlapping intervals. All data vectors were centered and normalized before being fed to the neural network. The solution of the classification problem includes subtasks. The first subtask was the classification for U.S. stock market and Russian stock market. The second subtask was the classification for U.S. stock market and the BRIC countries stock markets. The last subtask was the classification for stock markets in the BRIC countries group. Our results show the opportunity to split world stock markets into two groups. The U.S. constitutes the first group and the BRIC countries constitute the second group. Also, we can distinguish two separate groups for the BRIC countries stock markets. Brazil, Russia and India constitute the first group, and China constitutes the second group.</p>