Application of artificial intelligence in capital markets

The objective of this master thesis is to examine utilization of the neural network method of artificial intelligence to predict time series development within stock markets. Three neural network models were formed, from which the first one represents the recurrent LSTM neural network, formed using the Matlab software. The remaining two neural network models represent a feed-forward neural network, where one of them was based on the NAR model and the second one on the NARX model. Likewise, the two models were formed via the Matlab platform. There are seven stock indexes utilized as data inputs, namely three indexes from the USA, two from Europe and two from Asia. In order to evaluate the prediction capabilities of the examined models, statistical and profitability metrics are used. The recurrent LSTM network model reached positive profitability, contrary to the purchase strategy at the beginning of the investment period and to the sales at the end of the investment period. This applies to three out of the seven examined indexes. Nevertheless, the error rates between the real and the predicted development of stock indexes were very low, and therefore this model is evaluated as exploitable for the usage of prediction of examined stock indexes. The remaining two models were assessed in terms of the error rate and other statistical metrics, as they reached fine results in the selected period. Thus, they may be as well evaluated as suitable for prediction of the development of examined stock indexes. The principal objective of this thesis is to determine, whether it is possible to predict the future development of examined stock indexes with an adequate accuracy.