The usage of neural networks or other machine learning algorithms are prevalent in the sector of financial time series forecasting today. The mechanics governing the inner workings of these networks are mimicking the structure of naturally developed brains, in a simplified framework, but are developed to concentrate to one task and its underlying variables. These digital structures are referred to as universal approximators thanks to their abilities in different fields. The ForEx market and the stock exchanges are used as the school example to showcase Adam Smith’s perfect market conditions. My thesis is trying to investigate if cryptocurrency markets are even a better example of this by trying to measure the networks success on the ForEx and cryptocurrency markets. In this dissertation I introduce the aforementioned markets history and basic working principles and examine the modern artificial neural networks. I use simple statistical analysis to compare the results of these networks forecasts in order to estimate which one has more hidden variable, which one is harder to learn for an artificial neural network.